The intelligibility of demodulated audio signals from analog high frequency transmissions, e.g., using single-sideband (SSB) modulation, can be severely degraded by channel distortions and/or a mismatch between modulation and demodulation carrier frequency. In this work a neural network (NN)-based approach for carrier frequency offset (CFO) estimation from demodulated SSB signals is proposed, whereby a task specific architecture is presented. Additionally, a simulation framework for SSB signals is introduced and utilized for training the NNs. The CFO estimator is combined with a speech enhancement network to investigate its influence on the enhancement performance. The NN-based system is compared to a recently proposed pitch tracking based approach on publicly available data from real high frequency transmissions. Experiments show that the NN exhibits good CFO estimation properties and results in significant improvements in speech intelligibility, especially when combined with a noise reduction network.

Recent speaker diarization studies showed that integration of end-to-end neural diarization (EEND) and clustering-based diarization is a promising approach for achieving state-of-the-art performance on various tasks. Such an approach first divides an observed signal into fixed-length segments, then performs {\it segment-level} local diarization based on an EEND module, and merges the segment-level results via clustering to form a final global diarization result. The segmentation is done to limit the number of speakers in each segment since the current EEND cannot handle a large number of speakers. In this paper, we argue that such an approach involving the segmentation has several issues; for example, it inevitably faces a dilemma that larger segment sizes increase both the context available for enhancing the performance and the number of speakers for the local EEND module to handle. To resolve such a problem, this paper proposes a novel framework that performs diarization without segmentation. However, it can still handle challenging data containing many speakers and a significant amount of overlapping speech. The proposed method can take an entire meeting for inference and perform {\it utterance-by-utterance} diarization that clusters utterance activities in terms of speakers. To this end, we leverage a neural network training scheme called Graph-PIT proposed recently for neural source separation. Experiments with simulated active-meeting-like data and CALLHOME data show the superiority of the proposed approach over the conventional methods.

The scope of speech enhancement has changed from a monolithic view of single, independent tasks, to a joint processing of complex conversational speech recordings. Training and evaluation of these single tasks requires synthetic data with access to intermediate signals that is as close as possible to the evaluation scenario. As such data often is not available, many works instead use specialized databases for the training of each system component, e.g WSJ0-mix for source separation. We present a Multi-purpose Multi-Speaker Mixture Signal Generator (MMS-MSG) for generating a variety of speech mixture signals based on any speech corpus, ranging from classical anechoic mixtures (e.g., WSJ0-mix) over reverberant mixtures (e.g., SMS-WSJ) to meeting-style data. Its highly modular and flexible structure allows for the simulation of diverse environments and dynamic mixing, while simultaneously enabling an easy extension and modification to generate new scenarios and mixture types. These meetings can be used for prototyping, evaluation, or training purposes. We provide example evaluation data and baseline results for meetings based on the WSJ corpus. Further, we demonstrate the usefulness for realistic scenarios by using MMS-MSG to provide training data for the LibriCSS database.

Impressive progress in neural network-based single-channel speech source separation has been made in recent years. But those improvements have been mostly reported on anechoic data, a situation that is hardly met in practice. Taking the SepFormer as a starting point, which achieves state-of-the-art performance on anechoic mixtures, we gradually modify it to optimize its performance on reverberant mixtures. Although this leads to a word error rate improvement by 7 percentage points compared to the standard SepFormer implementation, the system ends up with only marginally better performance than a PIT-BLSTM separation system, that is optimized with rather straightforward means. This is surprising and at the same time sobering, challenging the practical usefulness of many improvements reported in recent years for monaural source separation on nonreverberant data.

In this work we address disentanglement of style and content in speech signals. We propose a fully convolutional variational autoencoder employing two encoders: a content encoder and a style encoder. To foster disentanglement, we propose adversarial contrastive predictive coding. This new disentanglement method does neither need parallel data nor any supervision. We show that the proposed technique is capable of separating speaker and content traits into the two different representations and show competitive speaker-content disentanglement performance compared to other unsupervised approaches. We further demonstrate an increased robustness of the content representation against a train-test mismatch compared to spectral features, when used for phone recognition.

In this paper we present our system for the Detection and Classification of Acoustic Scenes and Events (DCASE) 2021 Challenge Task 4: Sound Event Detection and Separation in Domestic Environments, where it scored the fourth rank. Our presented solution is an advancement of our system used in the previous edition of the task.We use a forward-backward convolutional recurrent neural network (FBCRNN) for tagging and pseudo labeling followed by tag-conditioned sound event detection (SED) models which are trained using strong pseudo labels provided by the FBCRNN. Our advancement over our earlier model is threefold. First, we introduce a strong label loss in the objective of the FBCRNN to take advantage of the strongly labeled synthetic data during training. Second, we perform multiple iterations of self-training for both the FBCRNN and tag-conditioned SED models. Third, while we used only tag-conditioned CNNs as our SED model in the previous edition we here explore sophisticated tag-conditioned SED model architectures, namely, bidirectional CRNNs and bidirectional convolutional transformer neural networks (CTNNs), and combine them. With metric and class specific tuning of median filter lengths for post-processing, our final SED model, consisting of 6 submodels (2 of each architecture), achieves on the public evaluation set poly-phonic sound event detection scores (PSDS) of 0.455 for scenario 1 and 0.684 for scenario as well as a collar-based F1-score of 0.596 outperforming the baselines and our model from the previous edition by far. Source code is publicly available at https://github.com/fgnt/pb_sed.

Recently, there has been a rising interest in sound recognition via Acoustic Sensor Networks to support applications such as ambient assisted living or environmental habitat monitoring. With state-of-the-art sound recognition being dominated by deep-learning-based approaches, there is a high demand for labeled training data. Despite the availability of large-scale data sets such as Google's AudioSet, acquiring training data matching a certain application environment is still often a problem. In this paper we are concerned with human activity monitoring in a domestic environment using an ASN consisting of multiple nodes each providing multichannel signals. We propose a self-training based domain adaptation approach, which only requires unlabeled data from the target environment. Here, a sound recognition system trained on AudioSet, the teacher, generates pseudo labels for data from the target environment on which a student network is trained. The student can furthermore glean information about the spatial arrangement of sensors and sound sources to further improve classification performance. It is shown that the student significantly improves recognition performance over the pre-trained teacher without relying on labeled data from the environment the system is deployed in.

Automatic transcription of meetings requires handling of overlapped speech, which calls for continuous speech separation (CSS) systems. The uPIT criterion was proposed for utterance-level separation with neural networks and introduces the constraint that the total number of speakers must not exceed the number of output channels. When processing meeting-like data in a segment-wise manner, i.e., by separating overlapping segments independently and stitching adjacent segments to continuous output streams, this constraint has to be fulfilled for any segment. In this contribution, we show that this constraint can be significantly relaxed. We propose a novel graph-based PIT criterion, which casts the assignment of utterances to output channels in a graph coloring problem. It only requires that the number of concurrently active speakers must not exceed the number of output channels. As a consequence, the system can process an arbitrary number of speakers and arbitrarily long segments and thus can handle more diverse scenarios. Further, the stitching algorithm for obtaining a consistent output order in neighboring segments is of less importance and can even be eliminated completely, not the least reducing the computational effort. Experiments on meeting-style WSJ data show improvements in recognition performance over using the uPIT criterion.

Speech activity detection (SAD), which often rests on the fact that the noise is "more'' stationary than speech, is particularly challenging in non-stationary environments, because the time variance of the acoustic scene makes it difficult to discriminate speech from noise. We propose two approaches to SAD, where one is based on statistical signal processing, while the other utilizes neural networks. The former employs sophisticated signal processing to track the noise and speech energies and is meant to support the case for a resource efficient, unsupervised signal processing approach. The latter introduces a recurrent network layer that operates on short segments of the input speech to do temporal smoothing in the presence of non-stationary noise. The systems are tested on the Fearless Steps challenge database, which consists of the transmission data from the Apollo-11 space mission. The statistical SAD achieves comparable detection performance to earlier proposed neural network based SADs, while the neural network based approach leads to a decision cost function of 1.07% on the evaluation set of the 2020 Fearless Steps Challenge, which sets a new state of the art.

We present an approach to deep neural network based (DNN-based) distance estimation in reverberant rooms for supporting geometry calibration tasks in wireless acoustic sensor networks. Signal diffuseness information from acoustic signals is aggregated via the coherent-to-diffuse power ratio to obtain a distance-related feature, which is mapped to a source-to-microphone distance estimate by means of a DNN. This information is then combined with direction-of-arrival estimates from compact microphone arrays to infer the geometry of the sensor network. Unlike many other approaches to geometry calibration, the proposed scheme does only require that the sampling clocks of the sensor nodes are roughly synchronized. In simulations we show that the proposed DNN-based distance estimator generalizes to unseen acoustic environments and that precise estimates of the sensor node positions are obtained.

In recent years time domain speech separation has excelled over frequency domain separation in single channel scenarios and noise-free environments. In this paper we dissect the gains of the time-domain audio separation network (TasNet) approach by gradually replacing components of an utterance-level permutation invariant training (u-PIT) based separation system in the frequency domain until the TasNet system is reached, thus blending components of frequency domain approaches with those of time domain approaches. Some of the intermediate variants achieve comparable signal-to-distortion ratio (SDR) gains to TasNet, but retain the advantage of frequency domain processing: compatibility with classic signal processing tools such as frequency-domain beamforming and the human interpretability of the masks. Furthermore, we show that the scale invariant signal-to-distortion ratio (si-SDR) criterion used as loss function in TasNet is related to a logarithmic mean square error criterion and that it is this criterion which contributes most reliable to the performance advantage of TasNet. Finally, we critically assess which gains in a noise-free single channel environment generalize to more realistic reverberant conditions.

The rising interest in single-channel multi-speaker speech separation sparked development of End-to-End (E2E) approaches to multispeaker speech recognition. However, up until now, state-of-theart neural network–based time domain source separation has not yet been combined with E2E speech recognition. We here demonstrate how to combine a separation module based on a Convolutional Time domain Audio Separation Network (Conv-TasNet) with an E2E speech recognizer and how to train such a model jointly by distributing it over multiple GPUs or by approximating truncated back-propagation for the convolutional front-end. To put this work into perspective and illustrate the complexity of the design space, we provide a compact overview of single-channel multi-speaker recognition systems. Our experiments show a word error rate of 11.0% on WSJ0-2mix and indicate that our joint time domain model can yield substantial improvements over cascade DNN-HMM and monolithic E2E frequency domain systems proposed so far.

Most approaches to multi-talker overlapped speech separation and recognition assume that the number of simultaneously active speakers is given, but in realistic situations, it is typically unknown. To cope with this, we extend an iterative speech extraction system with mechanisms to count the number of sources and combine it with a single-talker speech recognizer to form the first end-to-end multi-talker automatic speech recognition system for an unknown number of active speakers. Our experiments show very promising performance in counting accuracy, source separation and speech recognition on simulated clean mixtures from WSJ0-2mix and WSJ0-3mix. Among others, we set a new state-of-the-art word error rate on the WSJ0-2mix database. Furthermore, our system generalizes well to a larger number of speakers than it ever saw during training, as shown in experiments with the WSJ0-4mix database.

In this paper we present our system for the detection and classification of acoustic scenes and events (DCASE) 2020 Challenge Task 4: Sound event detection and separation in domestic environments. We introduce two new models: the forward-backward convolutional recurrent neural network (FBCRNN) and the tag-conditioned convolutional neural network (CNN). The FBCRNN employs two recurrent neural network (RNN) classifiers sharing the same CNN for preprocessing. With one RNN processing a recording in forward direction and the other in backward direction, the two networks are trained to jointly predict audio tags, i.e., weak labels, at each time step within a recording, given that at each time step they have jointly processed the whole recording. The proposed training encourages the classifiers to tag events as soon as possible. Therefore, after training, the networks can be applied to shorter audio segments of, e.g., 200ms, allowing sound event detection (SED). Further, we propose a tag-conditioned CNN to complement SED. It is trained to predict strong labels while using (predicted) tags, i.e., weak labels, as additional input. For training pseudo strong labels from a FBCRNN ensemble are used. The presented system scored the fourth and third place in the systems and teams rankings, respectively. Subsequent improvements allow our system to even outperform the challenge baseline and winner systems in average by, respectively, 18.0% and 2.2% event-based F1-score on the validation set. Source code is publicly available at https://github.com/fgnt/pb_sed.

Recently, the source separation performance was greatly improved by time-domain audio source separation based on dual-path recurrent neural network (DPRNN). DPRNN is a simple but effective model for a long sequential data. While DPRNN is quite efficient in modeling a sequential data of the length of an utterance, i.e., about 5 to 10 second data, it is harder to apply it to longer sequences such as whole conversations consisting of multiple utterances. It is simply because, in such a case, the number of time steps consumed by its internal module called inter-chunk RNN becomes extremely large. To mitigate this problem, this paper proposes a multi-path RNN (MPRNN), a generalized version of DPRNN, that models the input data in a hierarchical manner. In the MPRNN framework, the input data is represented at several (>_ 3) time-resolutions, each of which is modeled by a specific RNN sub-module. For example, the RNN sub-module that deals with the finest resolution may model temporal relationship only within a phoneme, while the RNN sub-module handling the most coarse resolution may capture only the relationship between utterances such as speaker information. We perform experiments using simulated dialogue-like mixtures and show that MPRNN has greater model capacity, and it outperforms the current state-of-the-art DPRNN framework especially in online processing scenarios.

We present an unsupervised training approach for a neural network-based mask estimator in an acoustic beamforming application. The network is trained to maximize a likelihood criterion derived from a spatial mixture model of the observations. It is trained from scratch without requiring any parallel data consisting of degraded input and clean training targets. Thus, training can be carried out on real recordings of noisy speech rather than simulated ones. In contrast to previous work on unsupervised training of neural mask estimators, our approach avoids the need for a possibly pre-trained teacher model entirely. We demonstrate the effectiveness of our approach by speech recognition experiments on two different datasets: one mainly deteriorated by noise (CHiME 4) and one by reverberation (REVERB). The results show that the performance of the proposed system is on par with a supervised system using oracle target masks for training and with a system trained using a model-based teacher.

We propose a training scheme to train neural network-based source separation algorithms from scratch when parallel clean data is unavailable. In particular, we demonstrate that an unsupervised spatial clustering algorithm is sufficient to guide the training of a deep clustering system. We argue that previous work on deep clustering requires strong supervision and elaborate on why this is a limitation. We demonstrate that (a) the single-channel deep clustering system trained according to the proposed scheme alone is able to achieve a similar performance as the multi-channel teacher in terms of word error rates and (b) initializing the spatial clustering approach with the deep clustering result yields a relative word error rate reduction of 26% over the unsupervised teacher.

Signal dereverberation using the Weighted Prediction Error (WPE) method has been proven to be an effective means to raise the accuracy of far-field speech recognition. First proposed as an iterative algorithm, follow-up works have reformulated it as a recursive least squares algorithm and therefore enabled its use in online applications. For this algorithm, the estimation of the power spectral density (PSD) of the anechoic signal plays an important role and strongly influences its performance. Recently, we showed that using a neural network PSD estimator leads to improved performance for online automatic speech recognition. This, however, comes at a price. To train the network, we require parallel data, i.e., utterances simultaneously available in clean and reverberated form. Here we propose to overcome this limitation by training the network jointly with the acoustic model of the speech recognizer. To be specific, the gradients computed from the cross-entropy loss between the target senone sequence and the acoustic model network output is backpropagated through the complex-valued dereverberation filter estimation to the neural network for PSD estimation. Evaluation on two databases demonstrates improved performance for on-line processing scenarios while imposing fewer requirements on the available training data and thus widening the range of applications.

In this paper, we present libDirectional, a MATLAB library for directional statistics and directional estimation. It supports a variety of commonly used distributions on the unit circle, such as the von Mises, wrapped normal, and wrapped Cauchy distributions. Furthermore, various distributions on higher-dimensional manifolds such as the unit hypersphere and the hypertorus are available. Based on these distributions, several recursive filtering algorithms in libDirectional allow estimation on these manifolds. The functionality is implemented in a clear, well-documented, and object-oriented structure that is both easy to use and easy to extend.

In this paper we present our audio tagging system for the DCASE 2019 Challenge Task 2. We propose a model consisting of a convolutional front end using log-mel-energies as input features, a recurrent neural network sequence encoder and a fully connected classifier network outputting an activity probability for each of the 80 considered event classes. Due to the recurrent neural network, which encodes a whole sequence into a single vector, our model is able to process sequences of varying lengths. The model is trained with only little manually labeled training data and a larger amount of automatically labeled web data, which hence suffers from label noise. To efficiently train the model with the provided data we use various data augmentation to prevent overfitting and improve generalization. Our best submitted system achieves a label-weighted label-ranking average precision (lwlrap) of 75.5% on the private test set which is an absolute improvement of 21.7% over the baseline. This system scored the second place in the teams ranking of the DCASE 2019 Challenge Task 2 and the fifth place in the Kaggle competition “Freesound Audio Tagging 2019” with more than 400 participants. After the challenge ended we further improved performance to 76.5% lwlrap setting a new state-of-the-art on this dataset.

In this paper we consider human daily activity recognition using an acoustic sensor network (ASN) which consists of nodes distributed in a home environment. Assuming that the ASN is permanently recording, the vast majority of recordings is silence. Therefore, we propose to employ a computationally efficient two-stage sound recognition system, consisting of an initial sound activity detection (SAD) and a subsequent sound event classification (SEC), which is only activated once sound activity has been detected. We show how a low-latency activity detector with high temporal resolution can be trained from weak labels with low temporal resolution. We further demonstrate the advantage of using spatial features for the subsequent event classification task.

Connectionist temporal classification (CTC) is a sequence-level loss that has been successfully applied to train recurrent neural network (RNN) models for automatic speech recognition. However, one major weakness of CTC is the conditional independence assumption that makes it difficult for the model to learn label dependencies. In this paper, we propose stimulated CTC, which uses stimulated learning to help CTC models learn label dependencies implicitly by using an auxiliary RNN to generate the appropriate stimuli. This stimuli comes in the form of an additional stimulation loss term which encourages the model to learn said label dependencies. The auxiliary network is only used during training and the inference model has the same structure as a standard CTC model. The proposed stimulated CTC model achieves about 35% relative character error rate improvements on a synthetic gesture keyboard recognition task and over 30% relative word error rate improvements on the Librispeech automatic speech recognition tasks over a baseline model trained with CTC only.

Despite the strong modeling power of neural network acoustic models, speech enhancement has been shown to deliver additional word error rate improvements if multi-channel data is available. However, there has been a longstanding debate whether enhancement should also be carried out on the ASR training data. In an extensive experimental evaluation on the acoustically very challenging CHiME-5 dinner party data we show that: (i) cleaning up the training data can lead to substantial error rate reductions, and (ii) enhancement in training is advisable as long as enhancement in test is at least as strong as in training. This approach stands in contrast and delivers larger gains than the common strategy reported in the literature to augment the training database with additional artificially degraded speech. Together with an acoustic model topology consisting of initial CNN layers followed by factorized TDNN layers we achieve with 41.6% and 43.2% WER on the DEV and EVAL test sets, respectively, a new single-system state-of-the-art result on the CHiME-5 data. This is a 8% relative improvement compared to the best word error rate published so far for a speech recognizer without system combination.

Multi-talker speech and moving speakers still pose a significant challenge to automatic speech recognition systems. Assuming an enrollment utterance of the target speakeris available, the so-called SpeakerBeam concept has been recently proposed to extract the target speaker from a speech mixture. If multi-channel input is available, spatial properties of the speaker can be exploited to support the source extraction. In this contribution we investigate different approaches to exploit such spatial information. In particular, we are interested in the question, how useful this information is if the target speaker changes his/her position. To this end, we present a SpeakerBeam-based source extraction network that is adapted to work on moving speakers by recursively updating the beamformer coefficients. Experimental results are presented on two data sets, one with articially created room impulse responses, and one with real room impulse responses and noise recorded in a conference room. Interestingly, spatial features turn out to be advantageous even if the speaker position changes.

This paper deals with multi-channel speech recognition in scenarios with multiple speakers. Recently, the spectral characteristics of a target speaker, extracted from an adaptation utterance, have been used to guide a neural network mask estimator to focus on that speaker. In this work we present two variants of speakeraware neural networks, which exploit both spectral and spatial information to allow better discrimination between target and interfering speakers. Thus, we introduce either a spatial preprocessing prior to the mask estimation or a spatial plus spectral speaker characterization block whose output is directly fed into the neural mask estimator. The target speaker’s spectral and spatial signature is extracted from an adaptation utterance recorded at the beginning of a session. We further adapt the architecture for low-latency processing by means of block-online beamforming that recursively updates the signal statistics. Experimental results show that the additional spatial information clearly improves source extraction, in particular in the same-gender case, and that our proposal achieves state-of-the-art performance in terms of distortion reduction and recognition accuracy.

In this paper, we present Hitachi and Paderborn University’s joint effort for automatic speech recognition (ASR) in a dinner party scenario. The main challenges of ASR systems for dinner party recordings obtained by multiple microphone arrays are (1) heavy speech overlaps, (2) severe noise and reverberation, (3) very natural onversational content, and possibly (4) insufficient training data. As an example of a dinner party scenario, we have chosen the data presented during the CHiME-5 speech recognition challenge, where the baseline ASR had a 73.3% word error rate (WER), and even the best performing system at the CHiME-5 challenge had a 46.1% WER. We extensively investigated a combination of the guided source separation-based speech enhancement technique and an already proposed strong ASR backend and found that a tight combination of these techniques provided substantial accuracy improvements. Our final system achieved WERs of 39.94% and 41.64% for the development and evaluation data, respectively, both of which are the best published results for the dataset. We also investigated with additional training data on the official small data in the CHiME-5 corpus to assess the intrinsic difficulty of this ASR task.

This paper presents an approach to voice conversion, whichdoes neither require parallel data nor speaker or phone labels fortraining. It can convert between speakers which are not in thetraining set by employing the previously proposed concept of afactorized hierarchical variational autoencoder. Here, linguisticand speaker induced variations are separated upon the notionthat content induced variations change at a much shorter timescale, i.e., at the segment level, than speaker induced variations,which vary at the longer utterance level. In this contribution wepropose to employ convolutional instead of recurrent networklayers in the encoder and decoder blocks, which is shown toachieve better phone recognition accuracy on the latent segmentvariables at frame-level due to their better temporal resolution.For voice conversion the mean of the utterance variables is re-placed with the respective estimated mean of the target speaker.The resulting log-mel spectra of the decoder output are used aslocal conditions of a WaveNet which is utilized for synthesis ofthe speech waveforms. Experiments show both good disentan-glement properties of the latent space variables, and good voiceconversion performance.

Automatic meeting analysis comprises the tasks of speaker counting, speaker diarization, and the separation of overlapped speech, followed by automatic speech recognition. This all has to be carried out on arbitrarily long sessions and, ideally, in an online or block-online manner. While significant progress has been made on individual tasks, this paper presents for the first time an all-neural approach to simultaneous speaker counting, diarization and source separation. The NN-based estimator operates in a block-online fashion and tracks speakers even if they remain silent for a number of time blocks, thus learning a stable output order for the separated sources. The neural network is recurrent over time as well as over the number of sources. The simulation experiments show that state of the art separation performance is achieved, while at the same time delivering good diarization and source counting results. It even generalizes well to an unseen large number of blocks.

In this paper we highlight the privacy risks entailed in deep neural network feature extraction for domestic activity monitoring. We employ the baseline system proposed in the Task 5 of the DCASE 2018 challenge and simulate a feature interception attack by an eavesdropper who wants to perform speaker identification. We then propose to reduce the aforementioned privacy risks by introducing a variational information feature extraction scheme that allows for good activity monitoring performance while at the same time minimizing the information of the feature representation, thus restricting speaker identification attempts. We analyze the resulting model’s composite loss function and the budget scaling factor used to control the balance between the performance of the trusted and attacker tasks. It is empirically demonstrated that the proposed method reduces speaker identification privacy risks without significantly deprecating the performance of domestic activity monitoring tasks.

Signal processing in WASNs is based on a software framework for hosting the algorithms as well as on a set of wireless connected devices representing the hardware. Each of the nodes contributes memory, processing power, communication bandwidth and some sensor information for the tasks to be solved on the network. In this paper we present our MARVELO framework for distributed signal processing. It is intended for transforming existing centralized implementations into distributed versions. To this end, the software only needs a block-oriented implementation, which MARVELO picks-up and distributes on the network. Additionally, our sensor node hardware and the audio interfaces responsible for multi-channel recordings are presented.

In this paper, we present a neural network based classification algorithm for the discrimination of moving from stationary targets in the sight of an automotive radar sensor. Compared to existing algorithms, the proposed algorithm can take into account multiple local radar targets instead of performing classification inference on each target individually resulting in superior discrimination accuracy, especially suitable for non rigid objects, like pedestrians, which in general have a wide velocity spread when multiple targets are detected.

Acoustic event detection, i.e., the task of assigning a human interpretable label to a segment of audio, has only recently attracted increased interest in the research community. Driven by the DCASE challenges and the availability of large-scale audio datasets, the state-of-the-art has progressed rapidly with deep-learning-based classi- fiers dominating the field. Because several potential use cases favor a realization on distributed sensor nodes, e.g. ambient assisted living applications, habitat monitoring or surveillance, we are concerned with two issues here. Firstly the classification performance of such systems and secondly the computing resources required to achieve a certain performance considering node level feature extraction. In this contribution we look at the balance between the two criteria by employing traditional techniques and different deep learning architectures, including convolutional and recurrent models in the context of real life everyday audio recordings in realistic, however challenging, multisource conditions.

Signal dereverberation using the weighted prediction error (WPE) method has been proven to be an effective means to raise the accuracy of far-field speech recognition. But in its original formulation, WPE requires multiple iterations over a sufficiently long utterance, rendering it unsuitable for online low-latency applications. Recently, two methods have been proposed to overcome this limitation. One utilizes a neural network to estimate the power spectral density (PSD) of the target signal and works in a block-online fashion. The other method relies on a rather simple PSD estimation which smoothes the observed PSD and utilizes a recursive formulation which enables it to work on a frame-by-frame basis. In this paper, we integrate a deep neural network (DNN) based estimator into the recursive frame-online formulation. We evaluate the performance of the recursive system with different PSD estimators in comparison to the block-online and offline variant on two distinct corpora. The REVERB challenge data, where the signal is mainly deteriorated by reverberation, and a database which combines WSJ and VoiceHome to also consider (directed) noise sources. The results show that although smoothing works surprisingly well, the more sophisticated DNN based estimator shows promising improvements and shortens the performance gap between online and offline processing.

Due to their distributed nature wireless acoustic sensor networks offer great potential for improved signal acquisition, processing and classification for applications such as monitoring and surveillance, home automation, or hands-free telecommunication. To reduce the communication demand with a central server and to raise the privacy level it is desirable to perform processing at node level. The limited processing and memory capabilities on a sensor node, however, stand in contrast to the compute and memory intensive deep learning algorithms used in modern speech and audio processing. In this work, we perform benchmarking of commonly used convolutional and recurrent neural network architectures on a Raspberry Pi based acoustic sensor node. We show that it is possible to run medium-sized neural network topologies used for speech enhancement and speech recognition in real time. For acoustic event recognition, where predictions in a lower temporal resolution are sufficient, it is even possible to run current state-of-the-art deep convolutional models with a real-time-factor of 0:11.

We present a block-online multi-channel front end for automatic speech recognition in noisy and reverberated environments. It is an online version of our earlier proposed neural network supported acoustic beamformer, whose coefficients are calculated from noise and speech spatial covariance matrices which are estimated utilizing a neural mask estimator. However, the sparsity of speech in the STFT domain causes problems for the initial beamformer coefficients estimation in some frequency bins due to lack of speech observations. We propose two methods to mitigate this issue. The first is to lower the frequency resolution of the STFT, which comes with the additional advantage of a reduced time window, thus lowering the latency introduced by block processing. The second approach is to smooth beamforming coefficients along the frequency axis, thus exploiting their high interfrequency correlation. With both approaches the gap between offline and block-online beamformer performance, as measured by the word error rate achieved by a downstream speech recognizer, is significantly reduced. Experiments are carried out on two copora, representing noisy (CHiME-4) and noisy reverberant (voiceHome) environments.

Distributed sensor data acquisition usually encompasses data sampling by the individual devices, where each of them has its own oscillator driving the local sampling process, resulting in slightly different sampling rates at the individual sensor nodes. Nevertheless, for certain downstream signal processing tasks it is important to compensate even for small sampling rate offsets. Aligning the sampling rates of oscillators which differ only by a few parts-per-million, is, however, challenging and quite different from traditional multirate signal processing tasks. In this paper we propose to transfer a precise but computationally demanding time domain approach, inspired by the Nyquist-Shannon sampling theorem, to an efficient frequency domain implementation. To this end a buffer control is employed which compensates for sampling offsets which are multiples of the sampling period, while a digital filter, realized by the wellknown Overlap-Save method, handles the fractional part of the sampling phase offset. With experiments on artificially misaligned data we investigate the parametrization, the efficiency, and the induced distortions of the proposed resampling method. It is shown that a favorable compromise between residual distortion and computational complexity is achieved, compared to other sampling rate offset compensation techniques.

It has been experimentally verified that sampling rate offsets (SROs) between the input channels of an acoustic beamformer have a detrimental effect on the achievable SNR gains. In this paper we derive an analytic model to study the impact of SRO on the estimation of the spatial noise covariance matrix used in MVDR beamforming. It is shown that a perfect compensation of the SRO is impossible if the noise covariance matrix is estimated by time averaging, even if the SRO is perfectly known. The SRO should therefore be compensated for prior to beamformer coefficient estimation. We present a novel scheme where SRO compensation and beamforming closely interact, saving some computational effort compared to separate SRO adjustment followed by acoustic beamforming.

The weighted prediction error (WPE) algorithm has proven to be a very successful dereverberation method for the REVERB challenge. Likewise, neural network based mask estimation for beamforming demonstrated very good noise suppression in the CHiME 3 and CHiME 4 challenges. Recently, it has been shown that this estimator can also be trained to perform dereverberation and denoising jointly. However, up to now a comparison of a neural beamformer and WPE is still missing, so is an investigation into a combination of the two. Therefore, we here provide an extensive evaluation of both and consequently propose variants to integrate deep neural network based beamforming with WPE. For these integrated variants we identify a consistent word error rate (WER) reduction on two distinct databases. In particular, our study shows that deep learning based beamforming benefits from a model-based dereverberation technique (i.e. WPE) and vice versa. Our key findings are: (a) Neural beamforming yields the lower WERs in comparison to WPE the more channels and noise are present. (b) Integration of WPE and a neural beamformer consistently outperforms all stand-alone systems.

NARA-WPE is a Python software package providing implementations of the weighted prediction error (WPE) dereverberation algorithm. WPE has been shown to be a highly effective tool for speech dereverberation, thus improving the perceptual quality of the signal and improving the recognition performance of downstream automatic speech recognition (ASR). It is suitable both for single-channel and multi-channel applications. The package consist of (1) a Numpy implementation which can easily be integrated into a custom Python toolchain, and (2) a TensorFlow implementation which allows integration into larger computational graphs and enables backpropagation through WPE to train more advanced front-ends. This package comprises of an iterative offline (batch) version, a block-online version, and a frame-online version which can be used in moderately low latency applications, e.g. digital speech assistants.

This paper describes the systems for the single-array track and the multiple-array track of the 5th CHiME Challenge. The final system is a combination of multiple systems, using Confusion Network Combination (CNC). The different systems presented here are utilizing different front-ends and training sets for a Bidirectional Long Short-Term Memory (BLSTM) Acoustic Model (AM). The front-end was replaced by enhancements provided by Paderborn University [1]. The back-end has been implemented using RASR [2] and RETURNN [3]. Additionally, a system combination including the hypothesis word graphs from the system of the submission [1] has been performed, which results in the final best system.

The invention of the Variational Autoencoder enables the application of Neural Networks to a wide range of tasks in unsupervised learning, including the field of Acoustic Unit Discovery (AUD). The recently proposed Hidden Markov Model Variational Autoencoder (HMMVAE) allows a joint training of a neural network based feature extractor and a structured prior for the latent space given by a Hidden Markov Model. It has been shown that the HMMVAE significantly outperforms pure GMM-HMM based systems on the AUD task. However, the HMMVAE cannot autonomously infer the number of acoustic units and thus relies on the GMM-HMM system for initialization. This paper introduces the Bayesian Hidden Markov Model Variational Autoencoder (BHMMVAE) which solves these issues by embedding the HMMVAE in a Bayesian framework with a Dirichlet Process Prior for the distribution of the acoustic units, and diagonal or full-covariance Gaussians as emission distributions. Experiments on TIMIT and Xitsonga show that the BHMMVAE is able to autonomously infer a reasonable number of acoustic units, can be initialized without supervision by a GMM-HMM system, achieves computationally efficient stochastic variational inference by using natural gradient descent, and, additionally, improves the AUD performance over the HMMVAE.

Deep clustering (DC) and deep attractor networks (DANs) are a data-driven way to monaural blind source separation. Both approaches provide astonishing single channel performance but have not yet been generalized to block-online processing. When separating speech in a continuous stream with a block-online algorithm, it needs to be determined in each block which of the output streams belongs to whom. In this contribution we solve this block permutation problem by introducing an additional speaker identification embedding to the DAN model structure. We motivate this model decision by analyzing the embedding topology of DC and DANs and show, that DC and DANs themselves are not sufficient for speaker identification. This model structure (a) improves the signal to distortion ratio (SDR) over a DAN baseline and (b) provides up to 61% and up to 34% relative reduction in permutation error rate and re-identification error rate compared to an i-vector baseline, respectively.

This contribution presents a speech enhancement system for the CHiME-5 Dinner Party Scenario. The front-end employs multi-channel linear time-variant filtering and achieves its gains without the use of a neural network. We present an adaptation of blind source separation techniques to the CHiME-5 database which we call Guided Source Separation (GSS). Using the baseline acoustic and language model, the combination of Weighted Prediction Error based dereverberation, guided source separation, and beamforming reduces the WER by 10:54% (relative) for the single array track and by 21:12% (relative) on the multiple array track.

Deep attractor networks (DANs) are a recently introduced method to blindly separate sources from spectral features of a monaural recording using bidirectional long short-term memory networks (BLSTMs). Due to the nature of BLSTMs, this is inherently not online-ready and resorting to operating on blocks yields a block permutation problem in that the index of each speaker may change between blocks. We here propose the joint modeling of spatial and spectral features to solve the block permutation problem and generalize DANs to multi-channel meeting recordings: The DAN acts as a spectral feature extractor for a subsequent model-based clustering approach. We first analyze different joint models in batch-processing scenarios and finally propose a block-online blind source separation algorithm. The efficacy of the proposed models is demonstrated on reverberant mixtures corrupted by real recordings of multi-channel background noise. We demonstrate that both the proposed batch-processing and the proposed block-online system outperform (a) a spatial-only model with a state-of-the-art frequency permutation solver and (b) a spectral-only model with an oracle block permutation solver in terms of signal to distortion ratio (SDR) gains.

This work examines acoustic beamformers employing neural networks (NNs) for mask prediction as front-end for automatic speech recognition (ASR) systems for practical scenarios like voice-enabled home devices. To test the versatility of the mask predicting network, the system is evaluated with different recording hardware, different microphone array designs, and different acoustic models of the downstream ASR system. Significant gains in recognition accuracy are obtained in all configurations despite the fact that the NN had been trained on mismatched data. Unlike previous work, the NN is trained on a feature level objective, which gives some performance advantage over a mask related criterion. Furthermore, different approaches for realizing online, or adaptive, NN-based beamforming are explored, where the online algorithms still show significant gains compared to the baseline performance.

This paper introduces a new open source platform for end-toend speech processing named ESPnet. ESPnet mainly focuses on end-to-end automatic speech recognition (ASR), and adopts widely-used dynamic neural network toolkits, Chainer and Py-Torch, as a main deep learning engine. ESPnet also follows the Kaldi ASR toolkit style for data processing, feature extraction/format, and recipes to provide a complete setup for speech recognition and other speech processing experiments. This paper explains a major architecture of this software platform, several important functionalities, which differentiate ESPnet from other open source ASR toolkits, and experimental results with major ASR benchmarks.

In this work, we address the limited availability of large annotated databases for real-life audio event detection by utilizing the concept of transfer learning. This technique aims to transfer knowledge from a source domain to a target domain, even if source and target have different feature distributions and label sets. We hypothesize that all acoustic events share the same inventory of basic acoustic building blocks and differ only in the temporal order of these acoustic units. We then construct a deep neural network with convolutional layers for extracting the acoustic units and a recurrent layer for capturing the temporal order. Under the above hypothesis, transfer learning from a source to a target domain with a different acoustic event inventory is realized by transferring the convolutional layers from the source to the target domain. The recurrent layer is, however, learnt directly from the target domain. Experiments on the transfer from a synthetic source database to the reallife target database of DCASE 2016 demonstrate that transfer learning leads to improved detection performance on average. However, the successful transfer to detect events which are very different from what was seen in the source domain, could not be verified.

In this paper we show how a neural network for spectral mask estimation for an acoustic beamformer can be optimized by algorithmic differentiation. Using the beamformer output SNR as the objective function to maximize, the gradient is propagated through the beamformer all the way to the neural network which provides the clean speech and noise masks from which the beamformer coefficients are estimated by eigenvalue decomposition. A key theoretical result is the derivative of an eigenvalue problem involving complex-valued eigenvectors. Experimental results on the CHiME-3 challenge database demonstrate the effectiveness of the approach. The tools developed in this paper are a key component for an end-to-end optimization of speech enhancement and speech recognition.

The benefits of both a logarithmic spectral amplitude (LSA) estimation and a modeling in a generalized spectral domain (where short-time amplitudes are raised to a generalized power exponent, not restricted to magnitude or power spectrum) are combined in this contribution to achieve a better tradeoff between speech quality and noise suppression in single-channel speech enhancement. A novel gain function is derived to enhance the logarithmic generalized spectral amplitudes of noisy speech. Experiments on the CHiME-3 dataset show that it outperforms the famous minimum mean squared error (MMSE) LSA gain function of Ephraim and Malah in terms of noise suppression by 1.4 dB, while the good speech quality of the MMSE-LSA estimator is maintained.

Recent advances in discriminatively trained mask estimation networks to extract a single source utilizing beamforming techniques demonstrate, that the integration of statistical models and deep neural networks (DNNs) are a promising approach for robust automatic speech recognition (ASR) applications. In this contribution we demonstrate how discriminatively trained embeddings on spectral features can be tightly integrated into statistical model-based source separation to separate and transcribe overlapping speech. Good generalization to unseen spatial configurations is achieved by estimating a statistical model at test time, while still leveraging discriminative training of deep clustering embeddings on a separate training set. We formulate an expectation maximization (EM) algorithm which jointly estimates a model for deep clustering embeddings and complex-valued spatial observations in the short time Fourier transform (STFT) domain at test time. Extensive simulations confirm, that the integrated model outperforms (a) a deep clustering model with a subsequent beamforming step and (b) an EM-based model with a beamforming step alone in terms of signal to distortion ratio (SDR) and perceptually motivated metric (PESQ) gains. ASR results on a reverberated dataset further show, that the aforementioned gains translate to reduced word error rates (WERs) even in reverberant environments.

Variational Autoencoders (VAEs) have been shown to provide efficient neural-network-based approximate Bayesian inference for observation models for which exact inference is intractable. Its extension, the so-called Structured VAE (SVAE) allows inference in the presence of both discrete and continuous latent variables. Inspired by this extension, we developed a VAE with Hidden Markov Models (HMMs) as latent models. We applied the resulting HMM-VAE to the task of acoustic unit discovery in a zero resource scenario. Starting from an initial model based on variational inference in an HMM with Gaussian Mixture Model (GMM) emission probabilities, the accuracy of the acoustic unit discovery could be significantly improved by the HMM-VAE. In doing so we were able to demonstrate for an unsupervised learning task what is well-known in the supervised learning case: Neural networks provide superior modeling power compared to GMMs.

In this paper, we apply a high-resolution approach, i.e. the matrix pencil method (MPM), to the FMCW automotive radar system to separate the neighboring targets, which share similar parameters, i.e. range, relative speed and azimuth angle, and cause overlapping in the radar spectrum. In order to adapt the 1D model of MPM to the 2D range-velocity spectrum and simultaneously limit the computational cost, some preprocessing steps are proposed to construct a novel separation algorithm. Finally, this algorithm is evaluated in both simulation and real data, and the results indicate a promising performance.

In this contribution we show how to exploit text data to support word discovery from audio input in an underresourced target language. Given audio, of which a certain amount is transcribed at the word level, and additional unrelated text data, the approach is able to learn a probabilistic mapping from acoustic units to characters and utilize it to segment the audio data into words without the need of a pronunciation dictionary. This is achieved by three components: an unsupervised acoustic unit discovery system, a supervisedly trained acoustic unit-to-grapheme converter, and a word discovery system, which is initialized with a language model trained on the text data. Experiments for multiple setups show that the initialization of the language model with text data improves the word segementation performance by a large margin.

In this paper, we present a hypothesis test for the classification of moving targets in the sight of an automotive radar sensor. For this purpose, a statistical model of the relative velocity between a stationary target and the radar sensor has been developed. With respect to the statistical properties a confidence interval is calculated and targets with relative velocity lying outside this interval are classified as moving targets. Compared to existing algorithms our approach is able to give robust classification independent of the number of observed moving targets and is characterized by an instantaneous classification, a simple parameterization of the model and an automatic calculation of the discriminating threshold.

In this paper we present an algorithm for the detection of moving targets in sight of an automotive radar sensor which can handle distorted ego-velocity information. In situations where biased or none velocity information is provided from the ego-vehicle, the algorithm is able to estimate the ego-velocity based on previously detected stationary targets with high accuracy, subsequently used for the target classification. Compared to existing ego-velocity algorithms our approach provides fast and efficient inference without sacrificing the practical classification accuracy. Other than that the algorithm is characterized by simple parameterization and little but appropriate model assumptions for high accurate production automotive radar sensors.

This paper presents an end-to-end training approach for a beamformer-supported multi-channel ASR system. A neural network which estimates masks for a statistically optimum beamformer is jointly trained with a network for acoustic modeling. To update its parameters, we propagate the gradients from the acoustic model all the way through feature extraction and the complex valued beamforming operation. Besides avoiding a mismatch between the front-end and the back-end, this approach also eliminates the need for stereo data, i.e., the parallel availability of clean and noisy versions of the signals. Instead, it can be trained with real noisy multichannel data only. Also, relying on the signal statistics for beamforming, the approach makes no assumptions on the configuration of the microphone array. We further observe a performance gain through joint training in terms of word error rate in an evaluation of the system on the CHiME 4 dataset.

Multi-channel speech enhancement algorithms rely on a synchronous sampling of the microphone signals. This, however, cannot always be guaranteed, especially if the sensors are distributed in an environment. To avoid performance degradation the sampling rate offset needs to be estimated and compensated for. In this contribution we extend the recently proposed coherence drift based method in two important directions. First, the increasing phase shift in the short-time Fourier transform domain is estimated from the coherence drift in a Matched Filterlike fashion, where intermediate estimates are weighted by their instantaneous SNR. Second, an observed bias is removed by iterating between offset estimation and compensation by resampling a couple of times. The effectiveness of the proposed method is demonstrated by speech recognition results on the output of a beamformer with and without sampling rate offset compensation between the input channels. We compare MVDR and maximum-SNR beamformers in reverberant environments and further show that both benefit from a novel phase normalization, which we also propose in this contribution.

In this contribution we investigate a priori signal-to-noise ratio (SNR) estimation, a crucial component of a single-channel speech enhancement system based on spectral subtraction. The majority of the state-of-the art a priori SNR estimators work in the power spectral domain, which is, however, not confirmed to be the optimal domain for the estimation. Motivated by the generalized spectral subtraction rule, we show how the estimation of the a priori SNR can be formulated in the so called generalized SNR domain. This formulation allows to generalize the widely used decision directed (DD) approach. An experimental investigation with different noise types reveals the superiority of the generalized DD approach over the conventional DD approach in terms of both the mean opinion score - listening quality objective measure and the output global SNR in the medium to high input SNR regime, while we show that the power spectrum is the optimal domain for low SNR. We further develop a parameterization which adjusts the domain of estimation automatically according to the estimated input global SNR. Index Terms: single-channel speech enhancement, a priori SNR estimation, generalized spectral subtraction

This contribution introduces a novel causal a priori signal-to-noise ratio (SNR) estimator for single-channel speech enhancement. To exploit the advantages of the generalized spectral subtraction, a normalized ?-order magnitude (NAOM) domain is introduced where an a priori SNR estimation is carried out. In this domain, the NAOM coefficients of noise and clean speech signals are modeled by a Weibull distribution and aWeibullmixturemodel (WMM), respectively. While the parameters of the noise model are calculated from the noise power spectral density estimates, the speechWMM parameters are estimated from the noisy signal by applying a causal Expectation-Maximization algorithm. Further a maximum a posteriori estimate of the a priori SNR is developed. The experiments in different noisy environments show the superiority of the proposed estimator compared to the well-known decision-directed approach in terms of estimation error, estimator variance and speech quality of the enhanced signals when used for speech enhancement.

A noise power spectral density (PSD) estimation is an indispensable component of speech spectral enhancement systems. In this paper we present a noise PSD tracking algorithm, which employs a noise presence probability estimate delivered by a deep neural network (DNN). The algorithm provides a causal noise PSD estimate and can thus be used in speech enhancement systems for communication purposes. An extensive performance comparison has been carried out with ten causal state-of-the-art noise tracking algorithms taken from the literature and categorized acc. to applied techniques. The experiments showed that the proposed DNN-based noise PSD tracker outperforms all competing methods with respect to all tested performance measures, which include the noise tracking performance and the performance of a speech enhancement system employing the noise tracking component.

Although complex-valued neural networks (CVNNs) â?? networks which can operate with complex arithmetic â?? have been around for a while, they have not been given reconsideration since the breakthrough of deep network architectures. This paper presents a critical assessment whether the novel tool set of deep neural networks (DNNs) should be extended to complex-valued arithmetic. Indeed, with DNNs making inroads in speech enhancement tasks, the use of complex-valued input data, specifically the short-time Fourier transform coefficients, is an obvious consideration. In particular when it comes to performing tasks that heavily rely on phase information, such as acoustic beamforming, complex-valued algorithms are omnipresent. In this contribution we recapitulate backpropagation in CVNNs, develop complex-valued network elements, such as the split-rectified non-linearity, and compare real- and complex-valued networks on a beamforming task. We find that CVNNs hardly provide a performance gain and conclude that the effort of developing the complex-valued counterparts of the building blocks of modern deep or recurrent neural networks can hardly be justified.

This paper is concerned with speech presence probability estimation employing an explicit model of the temporal and spectral correlations of speech. An undirected graphical model is introduced, based on a Factor Graph formulation. It is shown that this undirected model cures some of the theoretical issues of an earlier directed graphical model. Furthermore, we formulate a message passing inference scheme based on an approximate graph factorization, identify this inference scheme as a particular message passing schedule based on the turbo principle and suggest further alternative schedules. The experiments show an improved performance over speech presence probability estimation based on an IID assumption, and a slightly better performance of the turbo schedule over the alternatives.

This contribution investigates Direction of Arrival (DoA) estimation using linearly arranged microphone arrays. We are going to develop a model for the DoA estimation error in a reverberant scenario and show the existence of a bias, that is a consequence of the linear arrangement and limited field of view (FoV) bias: First, the limited FoV leading to a clipping of the measurements, and, second, the angular distribution of the signal energy of the reflections being non-uniform. Since both issues are a consequence of the linear arrangement of the sensors, the bias arises largely independent of the kind of DoA estimator. The experimental evaluation demonstrates the existence of the bias for a selected number of DoA estimation methods and proves that the prediction from the developed theoretical model matches the simulation results.

We present a system for the 4th CHiME challenge which significantly increases the performance for all three tracks with respect to the provided baseline system. The front-end uses a bi-directional Long Short-Term Memory (BLSTM)-based neural network to estimate signal statistics. These then steer a Generalized Eigenvalue beamformer. The back-end consists of a 22 layer deep Wide Residual Network and two extra BLSTM layers. Working on a whole utterance instead of frames allows us to refine Batch-Normalization. We also train our own BLSTM-based language model. Adding a discriminative speaker adaptation leads to further gains. The final system achieves a word error rate on the six channel real test data of 3.48%. For the two channel track we achieve 5.96% and for the one channel track 9.34%. This is the best reported performance on the challenge achieved by a single system, i.e., a configuration, which does not combine multiple systems. At the same time, our system is independent of the microphone configuration. We can thus use the same components for all three tracks.

In this paper we study the influence of directional radio patterns of Bluetooth low energy (BLE) beacons on smartphone localization accuracy and beacon network planning. A two-dimensional model of the power emission characteristic is derived from measurements of the radiation pattern of BLE beacons carried out in an RF chamber. The Cramer-Rao lower bound (CRLB) for position estimation is then derived for this directional power emission model. With this lower bound on the RMS positioning error the coverage of different beacon network configurations can be evaluated. For near-optimal network planing an evolutionary optimization algorithm for finding the best beacon placement is presented.

This paper describes automatic speech recognition (ASR) systems developed jointly by RWTH, UPB and FORTH for the 1ch, 2ch and 6ch track of the 4th CHiME Challenge. In the 2ch and 6ch tracks the final system output is obtained by a Confusion Network Combination (CNC) of multiple systems. The Acoustic Model (AM) is a deep neural network based on Bidirectional Long Short-Term Memory (BLSTM) units. The systems differ by front ends and training sets used for the acoustic training. The model for the 1ch track is trained without any preprocessing. For each front end we trained and evaluated individual acoustic models. We compare the ASR performance of different beamforming approaches: a conventional superdirective beamformer [1] and an MVDR beamformer as in [2], where the steering vector is estimated based on [3]. Furthermore we evaluated a BLSTM supported Generalized Eigenvalue beamformer using NN-GEV [4]. The back end is implemented using RWTH?s open-source toolkits RASR [5], RETURNN [6] and rwthlm [7]. We rescore lattices with a Long Short-Term Memory (LSTM) based language model. The overall best results are obtained by a system combination that includes the lattices from the system of UPB?s submission [8]. Our final submission scored second in each of the three tracks of the 4th CHiME Challenge.

In this paper we demonstrate an algorithm to learn words from speech using non-parametric Bayesian hierarchical models in an unsupervised setting. We exploit the assumption of a hierarchical structure of speech, namely the formation of spoken words as a sequence of phonemes. We employ the Nested Hierarchical Pitman-Yor Language Model, which allows an a priori unknown and possibly unlimited number of words. We assume the n-gram probabilities of words, the m-gram probabilities of phoneme sequences in words and the phoneme sequences of the words themselves as latent variables to be learned. We evaluate the algorithm on a cross language task using an existing speech recognizer trained on English speech to decode speech in the Xitsonga language supplied for the 2015 ZeroSpeech challenge. We apply the learning algorithm on the resulting phoneme graphs and achieve the highest token precision and F score compared to present systems.

Noise tracking is an important component of speech enhancement algorithms. Of the many noise trackers proposed, Minimum Statistics (MS) is a particularly popular one due to its simple parameterization and at the same time excellent performance. In this paper we propose to further reduce the number of MS parameters by giving an alternative derivation of an optimal smoothing constant. At the same time the noise tracking performance is improved as is demonstrated by experiments employing speech degraded by various noise types and at different SNR values.

We present a semantic analysis technique for spoken input using Markov Logic Networks (MLNs). MLNs combine graphical models with first-order logic. They areparticularly suitable for providing inference in the presence of inconsistent and incomplete data, which are typical of an automatic speech recognizer's (ASR) output in the presence of degraded speech. The target application is a speech interface to a home automation system to be operated by people with speech impairments, where the ASR output is particularly noisy. In order to cater for dysarthric speech with non-canonical phoneme realizations, acoustic representations of the input speech are learned in an unsupervised fashion. While training data transcripts are not required for the acoustic model training, the MLN training requires supervision, however, at a rather loose and abstract level. Results on two databases, one of them for dysarthric speech, show that MLN-based semantic analysis clearly outperforms baseline approaches employing non-negative matrix factorization, multinomial naive Bayes models, or support vector machines.

This contribution presents a Direction of Arrival (DoA) estimation algorithm based on the complex Watson distribution to incorporate both phase and level differences of captured micro- phone array signals. The derived algorithm is reviewed in the context of the Generalized State Coherence Transform (GSCT) on the one hand and a kernel density estimation method on the other hand. A thorough simulative evaluation yields insight into parameter selection and provides details on the performance for both directional and omni-directional microphones. A comparison to the well known Steered Response Power with Phase Transform (SRP-PHAT) algorithm and a state of the art DoA estimator which explicitly accounts for aliasing, shows in particular the advantages of presented algorithm if inter-sensor level differences are indicative of the DoA, as with directional microphones.

The parametric Bayesian Feature Enhancement (BFE) and a datadriven Denoising Autoencoder (DA) both bring performance gains in severe single-channel speech recognition conditions. The first can be adjusted to different conditions by an appropriate parameter setting, while the latter needs to be trained on conditions similar to the ones expected at decoding time, making it vulnerable to a mismatch between training and test conditions. We use a DNN backend and study reverberant ASR under three types of mismatch conditions: different room reverberation times, different speaker to microphone distances and the difference between artificially reverberated data and the recordings in a reverberant environment. We show that for these mismatch conditions BFE can provide the targets for a DA. This unsupervised adaptation provides a performance gain over the direct use of BFE and even enables to compensate for the mismatch of real and simulated reverberant data.

Only a few studies exist on automatic emotion analysis of speech from children with Autism Spectrum Conditions (ASC). Out of these, some preliminary studies have recently focused on comparing the relevance of selected prosodic features against large sets of acoustic, spectral, and cepstral features; however, no study so far provided a comparison of performances across different languages. The present contribution aims to fill this white spot in the literature and provide insight by extensive evaluations carried out on three databases of prompted phrases collected in English, Swedish, and Hebrew, inducing nine emotion categories embedded in short-stories. The datasets contain speech of children with ASC and typically developing children under the same conditions. We evaluate automatic diagnosis and recognition of emotions in atypical childrens voice over the nine categories including binary valence/arousal discrimination.

In this paper we present a source counting algorithm to determine the number of speakers in a speech mixture. In our proposed method, we model the histogram of estimated directions of arrival with a nonparametric Bayesian infinite Gaussian mixture model. As an alternative to classical model selection criteria and to avoid specifying the maximum number of mixture components in advance, a Dirichlet process prior is employed over the mixture components. This allows to automatically determine the optimal number of mixture components that most probably model the observations. We demonstrate by experiments that this model outperforms a parametric approach using a finite Gaussian mixture model with a Dirichlet distribution prior over the mixture weights.

"A method for nonstationary noise robust automatic speech recognition (ASR) is to first estimate the changing noise statistics and second clean up the features prior to recognition accordingly. Here, the first is accomplished by noise tracking in the spectral domain, while the second relies on Bayesian enhancement in the feature domain. In this way we take advantage of our recently proposed maximum a-posteriori based (MAP-B) noise power spectral density estimation algorithm, which is able to estimate the noise statistics even in time-frequency bins dominated by speech. We show that MAP-B noise tracking leads to an improved noise model estimate in the feature domain compared to estimating noise in speech absence periods only, if the bias resulting from the nonlinear transformation from the spectral to the feature domain is accounted for. Consequently, ASR results are improved, as is shown by experiments conducted on the Aurora IV database."

"In this contribution we derive a variational EM (VEM) algorithm for model selection in complex Watson mixture models, which have been recently proposed as a model of the distribution of normalized microphone array signals in the short-time Fourier transform domain. The VEM algorithm is applied to count the number of active sources in a speech mixture by iteratively estimating the mode vectors of the Watson distributions and suppressing the signals from the corresponding directions. A key theoretical contribution is the derivation of the MMSE estimate of a quadratic form involving the mode vector of the Watson distribution. The experimental results demonstrate the effectiveness of the source counting approach at moderately low SNR. It is further shown that the VEM algorithm is more robust w.r.t. used threshold values."

This contribution describes a step-wise source counting algorithm to determine the number of speakers in an offline scenario. Each speaker is identified by a variational expectation maximization (VEM) algorithm for complex Watson mixture models and therefore directly yields beamforming vectors for a subsequent speech separation process. An observation selection criterion is proposed which improves the robustness of the source counting in noise. The algorithm is compared to an alternative VEM approach with Gaussian mixture models based on directions of arrival and shown to deliver improved source counting accuracy. The article concludes by extending the offline algorithm towards a low-latency online estimation of the number of active sources from the streaming input data.

"In this paper we present an algorithm for the unsupervised segmentation of a lattice produced by a phoneme recognizer into words. Using a lattice rather than a single phoneme string accounts for the uncertainty of the recognizer about the true label sequence. An example application is the discovery of lexical units from the output of an error-prone phoneme recognizer in a zero-resource setting, where neither the lexicon nor the language model (LM) is known. We propose a computationally efficient iterative approach, which alternates between the following two steps: First, the most probable string is extracted from the lattice using a phoneme LM learned on the segmentation result of the previous iteration. Second, word segmentation is performed on the extracted string using a word and phoneme LM which is learned alongside the new segmentation. We present results on lattices produced by a phoneme recognizer on the WSJCAM0 dataset. We show that our approach delivers superior segmentation performance than an earlier approach found in the literature, in particular for higher-order language models. "

"Several self-localization algorithms have been proposed, that determine the positions of either acoustic or visual sensors autonomously. Usually these positions are given in a modality specific coordinate system, with an unknown rotation, translation and scale between the different systems. For a joint audiovisual tracking, where the different modalities support each other, the two modalities need to be mapped into a common coordinate system. In this paper we propose to estimate this mapping based on audiovisual correlates, i.e., a speaker that can be localized by both, a microphone and a camera network separately. The voice is tracked by a microphone network, which had to be calibrated by a self-localization algorithm at first, and the head is tracked by a calibrated camera network. Unlike existing Singular Value Decomposition based approaches to estimate the coordinate system mapping, we propose to perform an estimation in the shape domain, which turns out to be computationally more efficient. Simulations of the self-localization of an acoustic sensor network and a following coordinate mapping for a joint speaker localization showed a significant improvement of the localization performance, since the modalities were able to support each other."

"In this paper we present an approach for synchronizing the sampling clocks of distributed microphones over a wireless network. The proposed system uses a two stage procedure. It first employs a two-way message exchange algorithm to estimate the clock phase and frequency difference between two nodes and then uses a gossiping algorithmto estimate a virtual master clock, to which all sensor nodes synchronize. Simulation results are presented for networks of different topology and size, showing the effectiveness of our approach."

"Acoustic sensor network clock synchronization via time stamp exchange between the sensor nodes is not accurate enough for many acoustic signal processing tasks, such as speaker localization. To improve synchronization accuracy it has therefore been proposed to employ a Kalman Filter to obtain improved frequency deviation and phase offset estimates. The estimation requires a statistical model of the errors of the measurements obtained from the time stamp exchange algorithm. These errors are caused by random transmission delays and hardware effects and are thus network specific. In this contribution we develop an algorithm to estimate the parameters of the measurement error model alongside the Kalman filter based sampling clock synchronization, employing the Expectation Maximization algorithm. Simulation results demonstrate that the online estimation of the error model parameters leads only to a small degradation of the synchronization performance compared to a perfectly known observation error model."

In this paper, we investigate unsupervised acoustic model training approaches for dysarthric-speech recognition. These models are first, frame-based Gaussian posteriorgrams, obtained from Vector Quantization (VQ), second, so-called Acoustic Unit Descriptors (AUDs), which are hidden Markov models of phone-like units, that are trained in an unsupervised fashion, and, third, posteriorgrams computed on the AUDs. Experiments were carried out on a database collected from a home automation task and containing nine speakers, of which seven are considered to utter dysarthric speech. All unsupervised modeling approaches delivered significantly better recognition rates than a speaker-independent phoneme recognition baseline, showing the suitability of unsupervised acoustic model training for dysarthric speech. While the AUD models led to the most compact representation of an utterance for the subsequent semantic inference stage, posteriorgram-based representations resulted in higher recognition rates, with the Gaussian posteriorgram achieving the highest slot filling F-score of 97.02%. Index Terms: unsupervised learning, acoustic unit descriptors, dysarthric speech, non-negative matrix factorization

The accuracy of automatic speech recognition systems in noisy and reverberant environments can be improved notably by exploiting the uncertainty of the estimated speech features using so-called uncertainty-of-observation techniques. In this paper, we introduce a new Bayesian decision rule that can serve as a mathematical framework from which both known and new uncertainty-of-observation techniques can be either derived or approximated. The new decision rule in its direct form leads to the new significance decoding approach for Gaussian mixture models, which results in better performance compared to standard uncertainty-of-observation techniques in different additive and convolutive noise scenarios.

In this contribution we derive the Maximum A-Posteriori (MAP) estimates of the parameters of a Gaussian Mixture Model (GMM) in the presence of noisy observations. We assume the distortion to be white Gaussian noise of known mean and variance. An approximate conjugate prior of the GMM parameters is derived allowing for a computationally efficient implementation in a sequential estimation framework. Simulations on artificially generated data demonstrate the superiority of the proposed method compared to the Maximum Likelihood technique and to the ordinary MAP approach, whose estimates are corrected by the known statistics of the distortion in a straightforward manner.

In this paper we present an improved version of the recently proposed Maximum A-Posteriori (MAP) based noise power spectral density estimator. An empirical bias compensation and bandwidth adjustment reduce bias and variance of the noise variance estimates. The main advantage of the MAP-based postprocessor is its low estimation variance. The estimator is employed in the second stage of a two-stage single-channel speech enhancement system, where eight different state-of-the-art noise tracking algorithms were tested in the first stage. While the postprocessor hardly affects the results in stationary noise scenarios, it becomes the more effective the more nonstationary the noise is. The proposed postprocessor was able to improve all systems in babble noise w.r.t. the perceptual evaluation of speech quality performance.

Among the different configurations of multi-microphone systems, e.g., in applications of speech dereverberation or denoising, we consider the case without a priori information of the microphone-array geometry. This naturally invokes explicit or implicit identification of source-receiver transfer functions as an indirect description of the microphone-array configuration. However, this blind channel identification (BCI) has been difficult due to the lack of unique identifiability in the presence of observation noise or near-common channel zeros. In this paper, we study the implicit BCI performance of blind signal enhancement techniques such as the adaptive principal component analysis (PCA) or the iterative blind equalization and channel identification (BENCH). To this end, we make use of a recently proposed metric, the normalized filter-projection misalignment (NFPM), which is tailored for BCI evaluation in ill-conditioned (e.g., noisy) scenarios. The resulting understanding of implicit BCI performance can help to judge the behavior of multi-microphone speech enhancement systems and the suitability of implicit BCI to serve channel-based (i.e., channel-informed) enhancement.

In this paper, we consider the Maximum Likelihood (ML) estimation of the parameters of a GAUSSIAN in the presence of censored, i.e., clipped data. We show that the resulting Expectation Maximization (EM) algorithm delivers virtually biasfree and efficient estimates, and we discuss its convergence properties. We also discuss optimal classification in the presence of censored data. Censored data are frequently encountered in wireless LAN positioning systems based on the fingerprinting method employing signal strength measurements, due to the limited sensitivity of the portable devices. Experiments both on simulated and real-world data demonstrate the effectiveness of the proposed algorithms.

In this paper we present a modified hidden Markov model (HMM) for the fusion of received signal strength index (RSSI) information of WiFi access points and relative position information which is obtained from the inertial sensors of a smartphone for indoor positioning. Since the states of the HMM represent the potential user locations, their number determines the quantization error introduced by discretizing the allowable user positions through the use of the HMM. To reduce this quantization error we introduce â??pseudoâ?? states, whose emission probability, which models the RSSI measurements at this location, is synthesized from those of the neighboring states of which a Gaussian emission probability has been estimated during the training phase. The experimental results demonstrate the effectiveness of this approach. By introducing on average two pseudo states per original HMM state the positioning error could be significantly reduced without increasing the training effort.

In this paper we present a system for indoor navigation based on received signal strength index information of Wireless-LAN access points and relative position estimates. The relative position information is gathered from inertial smartphone sensors using a step detection and an orientation estimate. Our map data is hosted on a server employing a map renderer and a SQL database. The database includes a complete multilevel office building, within which the user can navigate. During navigation, the client retrieves the position estimate from the server, together with the corresponding map tiles to visualize the user's position on the smartphone display.

In this paper we propose an approach to retrieve the absolute geometry of an acoustic sensor network, consisting of spatially distributed microphone arrays, from reverberant speech input. The calibration relies on direction of arrival measurements of the individual arrays. The proposed calibration algorithm is derived from a maximum-likelihood approach employing circular statistics. Since a sensor node consists of a microphone array with known intra-array geometry, we are able to obtain an absolute geometry estimate, including angles and distances. Simulation results demonstrate the effectiveness of the approach.

Recently, substantial progress has been made in the field of reverberant speech signal processing, including both single- and multichannel de-reverberation techniques, and automatic speech recognition (ASR) techniques robust to reverberation. To evaluate state-of-the-art algorithms and obtain new insights regarding potential future research directions, we propose a common evaluation framework including datasets, tasks, and evaluation metrics for both speech enhancement and ASR techniques. The proposed framework will be used as a common basis for the REVERB (REverberant Voice Enhancement and Recognition Benchmark) challenge. This paper describes the rationale behind the challenge, and provides a detailed description of the evaluation framework and benchmark results.

In this paper we present a combined hardware/software approach for synchronizing the sampling clocks of an acoustic sensor network. A first clock frequency offset estimate is obtained by a time stamp exchange protocol with a low data rate and computational requirements. The estimate is then postprocessed by a Kalman filter which exploits the specific properties of the statistics of the frequency offset estimation error. In long term experiments the deviation between the sampling oscillators of two sensor nodes never exceeded half a sample with a wired and with a wireless link between the nodes. The achieved precision enables the estimation of time difference of arrival values across different hardware devices without sharing a common sampling hardware.

We present a novel method to exploit correlations of adjacent time-frequency (TF)-slots for a sparseness-based blind speech separation (BSS) system. Usually, these correlations are exploited by some heuristic smoothing techniques in the post-processing of the estimated soft TF masks. We propose a different approach: Based on our previous work with one-dimensional (1D)-hidden Markov models (HMMs) along the time axis we extend the modeling to two-dimensional (2D)-HMMs to exploit both temporal and spectral correlations in the speech signal. Based on the principles of turbo decoding we solved the complex inference of 2D-HMMs by a modified forward-backward algorithm which operates alternatingly along the time and the frequency axis. Extrinsic information is exchanged between these steps such that increasingly better soft time-frequency masks are obtained, leading to improved speech separation performance in highly reverberant recording conditions.

In this paper we present a speech presence probability (SPP) estimation algorithmwhich exploits both temporal and spectral correlations of speech. To this end, the SPP estimation is formulated as the posterior probability estimation of the states of a two-dimensional (2D) Hidden Markov Model (HMM). We derive an iterative algorithm to decode the 2D-HMM which is based on the turbo principle. The experimental results show that indeed the SPP estimates improve from iteration to iteration, and further clearly outperform another state-of-the-art SPP estimation algorithm.

In this paper we consider the unsupervised word discovery from phonetic input. We employ a word segmentation algorithm which simultaneously develops a lexicon, i.e., the transcription of a word in terms of a phone sequence, learns a n-gram language model describing word and word sequence probabilities, and carries out the segmentation itself. The underlying statistical model is that of a Pitman-Yor process, a concept known from Bayesian non-parametrics, which allows for an a priori unknown and unlimited number of different words. Using a hierarchy of Pitman-Yor processes, language models of different order can be employed and nesting it with another hierarchy of Pitman-Yor processes on the phone level allows for backing off unknown word unigrams by phone m-grams. We present results on a large-vocabulary task, assuming an error-free phone sequence is given. We finish by discussing options how to cope with noisy phone sequences.

In this paper we present a novel noise power spectral density tracking algorithm and its use in single-channel speech enhancement. It has the unique feature that it is able to track the noise statistics even if speech is dominant in a given time-frequency bin. As a consequence it can follow non-stationary noise superposed by speech, even in the critical case of rising noise power. The algorithm requires an initial estimate of the power spectrum of speech and is thus meant to be used as a postprocessor to a first speech enhancement stage. An experimental comparison with a state-of-the-art noise tracking algorithm demonstrates lower estimation errors under low SNR conditions and smaller fluctuations of the estimated values, resulting in improved speech quality as measured by PESQ scores.

In this paper we propose an approach to retrieve the geometry of an acoustic sensor network consisting of spatially distributed microphone arrays from unconstrained speech input. The calibration relies on Direction of Arrival (DoA) measurements which do not require a clock synchronization among the sensor nodes. The calibration problem is formulated as a cost function optimization task, which minimizes the squared differences between measured and predicted observations and additionally avoids the existence of minima that correspond to mirrored versions of the actual sensor orientations. Further, outlier measurements caused by reverberation are mitigated by a Random Sample Consensus (RANSAC) approach. The experimental results show a mean positioning error of at most 25 cm even in highly reverberant environments.

In this contribution we investigate the effectiveness of Bayesian feature enhancement (BFE) on a medium-sized recognition task containing real-world recordings of noisy reverberant speech. BFE employs a very coarse model of the acoustic impulse response (AIR) from the source to the microphone, which has been shown to be effective if the speech to be recognized has been generated by artificially convolving nonreverberant speech with a constant AIR. Here we demonstrate that the model is also appropriate to be used in feature enhancement of true recordings of noisy reverberant speech. On the Multi-Channel Wall Street Journal Audio Visual corpus (MC-WSJ-AV) the word error rate is cut in half to 41.9 percent compared to the ETSI Standard Front-End using as input the signal of a single distant microphone with a single recognition pass.

In this work, an observation model for the joint compensation of noise and reverberation in the logarithmic mel power spectral density domain is considered. It relates the features of the noisy reverberant speech to those of the non-reverberant speech and the noise. In contrast to enhancement of features only corrupted by reverberation (reverberant features), enhancement of noisy reverberant features requires a more sophisticated model for the error introduced by the proposed observation model. In a first consideration, it will be shown that this error is highly dependent on the instantaneous ratio of the power of reverberant speech to the power of the noise and, moreover, sensitive to the phase between reverberant speech and noise in the short-time discrete Fourier domain. Afterwards, a statistically motivated approach will be presented allowing for the model of the observation error to be inferred from the error model previously used for the reverberation only case. Finally, the developed observation error model will be utilized in a Bayesian feature enhancement scheme, leading to improvements in word accuracy on the AURORA5 database.

In this paper we present a system for car navigation by fusing sensor data on an Android smartphone. The key idea is to use both the internal sensors of the smartphone (e.g., gyroscope) and sensor data from the car (e.g., speed information) to support navigation via GPS. To this end we employ a CAN-Bus-to-Bluetooth adapter to establish a wireless connection between the smartphone and the CAN-Bus of the car. On the smartphone a strapdown algorithm and an error-state Kalman filter are used to fuse the different sensor data streams. The experimental results show that the system is able to maintain higher positioning accuracy during GPS dropouts, thus improving the availability and reliability, compared to GPS-only solutions.

In this paper we present our experimental results about classifying audio data into broad acoustic categories. The reverberated sound samples from indoor recordings are grouped into four classes, namely speech, music, acoustic events and noise. We investigated a total of 188 acoustic features and achieved for the best configuration a classification accuracy better than 98\%. This was achieved by a 42-dimensional feature vector consisting of Mel-Frequency Cepstral Coefficients, an autocorrelation feature and so-called track features that measure the length of ''traces'' of high energy in the spectrogram. We also found a 4-feature configuration with a classification rate of about 90\% allowing for broad acoustic category classification with low computational effort.

The paper proposes a modification of the standard maximum a posteriori (MAP) method for the estimation of the parameters of a Gaussian process for cases where the process is superposed by additive Gaussian observation errors of known variance. Simulations on artificially generated data demonstrate the superiority of the proposed method. While reducing to the ordinary MAP approach in the absence of observation noise, the improvement becomes the more pronounced the larger the variance of the observation noise. The method is further extended to track the parameters in case of non-stationary Gaussian processes.

In this work, a splitting and weighting scheme that allows for splitting a Gaussian density into a Gaussian mixture density (GMM) is extended to allow the mixture components to be arranged along arbitrary directions. The parameters of the Gaussian mixture are chosen such that the GMM and the original Gaussian still exhibit equal central moments up to an order of four. The resulting mixtures{\rq} covariances will have eigenvalues that are smaller than those of the covariance of the original distribution, which is a desirable property in the context of non-linear state estimation, since the underlying assumptions of the extended K ALMAN filter are better justified in this case. Application to speech feature enhancement in the context of noise-robust automatic speech recognition reveals the beneficial properties of the proposed approach in terms of a reduced word error rate on the Aurora 2 recognition task.

In this paper we propose to jointly consider Segmental Dynamic Time Warping and distance clustering for the unsupervised learning of acoustic events. As a result, the computational complexity increases only linearly with the dababase size compared to a quadratic increase in a sequential setup, where all pairwise SDTW distances between segments are computed prior to clustering. Further, we discuss options for seed value selection for clustering and show that drawing seeds with a probability proportional to the distance from the already drawn seeds, known as K-means++ clustering, results in a significantly higher probability of finding representatives of each of the underlying classes, compared to the commonly used draws from a uniform distribution. Experiments are performed on an acoustic event classification and an isolated digit recognition task, where on the latter the final word accuracy approaches that of supervised training.

In this paper we propose a procedure for estimating the geometric configuration of an arbitrary acoustic sensor placement. It determines the position and the orientation of microphone arrays in 2D while locating a source by direction-of-arrival (DoA) estimation. Neither artificial calibration signals nor unnatural user activity are required. The problem of scale indeterminacy inherent to DoA-only observations is solved by adding time difference of arrival (TDOA) measurements. The geometry calibration method is numerically stable and delivers precise results in moderately reverberated rooms. Simulation results are confirmed by laboratory experiments.

In this paper we address the problem of initial seed selection for frequency domain iterative blind speech separation (BSS) algorithms. The derivation of the seeding algorithm is guided by the goal to select samples which are likely to be caused by source activity and not by noise and at the same time originate from different sources. The proposed algorithm has moderate computational complexity and finds better seed values than alternative schemes, as is demonstrated by experiments on the database of the SiSEC2010 challenge.

In this paper we present a robust location estimation algorithm especially focused on the accuracy in vertical position. A loosely-coupled error state space Kalman filter, which fuses sensor data of an Inertial Measurement Unit and the output of a Global Positioning System device, is augmented by height information from an altitude measurement unit. This unit consists of a barometric altimeter whose output is fused with topographic map information by a Kalman filter to provide robust information about the current vertical user position. These data replace the less reliable vertical position information provided the GPS device. It is shown that typical barometric errors like thermal divergences and fluctuations in the pressure due to changing weather conditions can be compensated by the topographic map information and the barometric error Kalman filter. The resulting height information is shown not only to be more reliable than height information provided by GPS. It also turns out that it leads to better attitude and thus better overall localization estimation accuracy due to the coupling of spatial orientations via the Direct Cosine Matrix. Results are presented both for artificially generated and field test data, where the user is moving by car.

Traditionally, ASR systems are based on hidden Markov models with Gaussian mixtures modelling the state-conditioned feature distribution. The inherent assumption of conditional independence, stating that a feature's likelihood solely depends on the current HMM state, makes the search computationally tractable, nevertheless has also been identified to be a major reason for the lack of robustness of such systems. Linear dynamic models have been proposed to overcome this weakness by employing a hidden dynamic state process underlying the observed features. Though performance of linear dynamic models on continuous speech/phone recognition tasks has been shown to be superior to that of equivalent static models, this approach still cannot compete with the established acoustic models. In this paper we consider the combination of hidden Markov models based on Gaussian mixture densities (GMM-HMMs) and linear dynamic models (LDMs) as the acoustic model for automatic speech recognition systems. In doing so, the individual strengths of both models, i.e. the modelling of long-term temporal dependencies by the GMM-HMM and the direct modelling of statistical dependencies between consecutive feature vectors by the LDM, are exploited. Phone classification experiments conducted on the TIMIT database indicate the prospective use of this approach for the application to continuous speech recognition.

Linear dynamic models (LDMs) have been shown to be a viable alternative to hidden Markov models (HMMs) on small-vocabulary recognition tasks, such as phone classification. In this paper we investigate various statistical model combination approaches for a hybrid HMM-LDM recognizer, resulting in a phone classification performance that outperforms the best individual classifier. Further, we report on continuous speech recognition experiments on the AURORA4 corpus, where the model combination is carried out on wordgraph rescoring. While the hybrid system improves the HMM system in the case of monophone HMMs, the performance of the triphone HMM model could not be improved by monophone LDMs, asking for the need to introduce context-dependency also in the LDM model inventory.

We describe an algorithm that performs regularized non-negative matrix factorization (NMF) to find independent components in non-negative data. Previous techniques proposed for this purpose require the data to be grounded, with support that goes down to 0 along each dimension. In our work, this requirement is eliminated. Based on it, we present a technique to find a low-dimensional decomposition of spectrograms by casting it as a problem of discovering independent non-negative components from it. The algorithm itself is implemented as regularized non-negative matrix factorization (NMF). Unlike other ICA algorithms, this algorithm computes the mixing matrix rather than an unmixing matrix. This algorithm provides a better decomposition than standard NMF when the underlying sources are independent. It makes better use of additional observation streams than previous non-negative ICA algorithms.

In this contribution we provide a unified treatment of blind source separation (BSS) and noise suppression, two tasks which have traditionally been considered different and for which quite different techniques have been developed. Exploiting the sparseness of the sources in the short time frequency domain and using a probabilistic model which accounts for the presence of additive noise and which captures the spatial information of the multi-channel recording, a speech enhancement system is developed which suppresses noise and simultaneously separates speakers in case multiple speakers are active. Source activity estimation and model parameter estimation form the E-step and the M-step of the Expectation Maximization algorithm, respectively. Experimental results obtained on the dataset of the Signal Separation Evaluation Campaign 2010 demonstrate the effectiveness of the proposed system.

In this paper we propose to employ directional statistics in a complex vector space to approach the problem of blind speech separation in the presence of spatially correlated noise. We interpret the values of the short time Fourier transform of the microphone signals to be draws from a mixture of complex Watson distributions, a probabilistic model which naturally accounts for spatial aliasing. The parameters of the density are related to the a priori source probabilities, the power of the sources and the transfer function ratios from sources to sensors. Estimation formulas are derived for these parameters by employing the Expectation Maximization (EM) algorithm. The E-step corresponds to the estimation of the source presence probabilities for each time-frequency bin, while the M-step leads to a maximum signal-to-noise ratio (MaxSNR) beamformer in the presence of uncertainty about the source activity. Experimental results are reported for an implementation in a generalized sidelobe canceller (GSC) like spatial beamforming configuration for 3 speech sources with significant coherent noise in reverberant environments, demonstrating the usefulness of the novel modeling framework.

In this paper we present a novel vehicle tracking algorithm, which is based on multi-level sensor fusion of GPS (global positioning system) with Inertial Measurement Unit sensor data. It is shown that the robustness of the system to temporary dropouts of the GPS signal, which may occur due to limited visibility of satellites in narrow street canyons or tunnels, is greatly improved by sensor fusion. We further demonstrate how the observation and state noise covariances of the employed Kalman filters can be estimated alongside the filtering by an application of the Expectation-Maximization algorithm. The proposed time-variant multi-level Kalman filter is shown to outperform an Interacting Multiple Model approach while at the same time being computationally less demanding.

In this paper we present a novel vehicle tracking method which is based on multi-stage Kalman filtering of GPS and IMU sensor data. After individual Kalman filtering of GPS and IMU measurements the estimates of the orientation of the vehicle are combined in an optimal manner to improve the robustness towards drift errors. The tracking algorithm incorporates the estimation of time-variant covariance parameters by using an iterative block Expectation-Maximization algorithm to account for time-variant driving conditions and measurement quality. The proposed system is compared to an interacting multiple model approach (IMM) and achieves improved localization accuracy at lower computational complexity. Furthermore we show how the joint parameter estimation and localizaiton can be conducted with streaming input data to be able to track vehicles in a real driving environment.

Microphone arrays represent the basis for many challenging acoustic sensing tasks. The accuracy of techniques like beamforming directly depends on a precise knowledge of the relative positions of the sensors used. Unfortunately, for certain use cases manually measuring the geometry of an array is not feasible due to practical constraints. In this paper we present an approach to unsupervised shape calibration of microphone array networks. We developed a hierarchical procedure that first performs local shape calibration based on coherence analysis and then employs SRP-PHAT in a network calibration method. Practical experiments demonstrate the effectiveness of our approach especially for highly reverberant acoustic environments.

In this paper we present a new feature space dereverberation technique for automatic speech recognition. We derive an expression for the dependence of the reverberant speech features in the log-mel spectral domain on the non-reverberant speech features and the room impulse response. The obtained observation model is used for a model based speech enhancement based on Kalman filtering. The performance of the proposed enhancement technique is studied on the AURORA5 database. In our currently best configuration, which includes uncertainty decoding, the number of recognition errors is approximately halved compared to the recognition of unprocessed speech.

In this paper we present an Uncertainty Decoding rule which exploits feature reliability information and interframe correlation for noise robust speech recognition. The reliability information can be obtained either from conditional Bayesian estimation, where speech and noise feature vectors are tracked jointly, or by augmenting conventional point estimation methods with heuristics about the estimator's reliability. Experimental results on the AURORA2 database demonstrate on the one hand that Uncertainty Decoding improves recognition performance, while on the other hand it is seen that the severe approximations needed to arrive at computationally tractable solutions have their noticable impact on recognition performance.

In this paper we present an analytic derivation of the moments of the phase factor between clean speech and noise cepstral or log-mel-spectral feature vectors. The development shows, among others, that the probability density of the phase factor is of sub-Gaussian nature and that it is independent of the noise type and the signal-to-noise ratio, however dependent on the mel filter bank index. Further we show how to compute the contribution of the phase factor to both the mean and the vari- ance of the noisy speech observation likelihood, which relates the speech and noise feature vectors to those of noisy speech. The resulting phase-sensitive observation model is then used in model-based speech feature enhancement, leading to significant improvements in word accuracy on the AURORA2 database.

A combination of GPS (global positioning system) and INS (inertial navigation system) is known to provide high precision and highly robust vehicle localization. Notably during times when the GPS signal has a poor quality, e.g. due to the lack of a sufficiently large number of visible satellites, the INS, which may consist of a gyroscope and an odometer, will lead to improved positioning accuracy. In this paper we show how velocity information obtained from GSM (global system for mobile communications) signalling, rather than from a tachometer, can be used together with a gyroscope sensor to support localization in the presence of temporarily unavailable GPS data. We propose a sensor fusion system architecture and present simulation results that show the effectiveness of this approach.

In this paper we present a system for identifying and localizingspeakers using distant microphone arrays and a steerablepan-tilt-zoom camera. Audio and video streams are processedin real-time to obtain the diarization information {grqq}who speakswhen and where'' with low latency to be used in advanced videoconferencing systems or user-adaptive interfaces. A key featureof the proposed system is to first glean information about thespeaker{\rq}s location and identity from the audio and visual datastreams separately and then to fuse these data in a probabilisticframework employing the Viterbi algorithm. Here, visual evidenceof a person is utilized through a priori state probabilities,while location and speaker change information are employedvia time-variant transition probablities. Experiments show thatvideo information yields a substantial improvement comparedto pure audio-based diarization.

In diesem Beitrag werden zwei neuartige akustische Strahlformungsalgorithmen fuer einen Einsatz im KFZ diskutiert. In beiden Verfahren ist fuer jede Frequenzkomponente der Eigenvektor zum groessten Eigenwert eines verallgemeinerten Eigenwertproblems zu bestimmen: bei der einen Varianate stellen die Filterkoeffizienten gerade obigen Eigenvektor dar. Sprachverzerrungen werden mit einem einkanaligen Nachfilter kompensiert. In der anderen Variante, welche die Struktur eines ''Generalized Sidelobe Cancellers'' (GSC) hat, basiert die Blockiermatrix auf obiger Eigenvektorzerlegung.