Endpoint (EP) detection is a key component of far-field speech recognition systems that assist the user through voice commands. The endpoint detector has to trade-off between accuracy and latency, since waiting longer reduces the cases of users being cut-off early. We propose a novel two-pass solution for endpointing, where the utterance endpoint detected from a first pass endpointer is verified by a 2nd-pass model termed EP Arbitrator. Our method improves the trade-off between early cut-offs and latency over a baseline endpointer, as tested on datasets including voice-assistant transactional queries, conversational speech, and the public SLURP corpus. We demonstrate that our method shows improvements regardless of the first-pass EP model used.

We propose a novel approach for ASR N-best hypothesis rescoring with graph-based label propagation by leveraging cross-utterance acoustic similarity. In contrast to conventional neural language model (LM) based ASR rescoring/reranking models, our approach focuses on acoustic information and conducts the rescoring collaboratively among utterances, instead of individually. Experiments on the VCTK dataset demonstrate that our approach consistently improves ASR performance, as well as fairness across speaker groups with different accents. Our approach provides a low-cost solution for mitigating the majoritarian bias of ASR systems, without the need to train new domain- or accent-specific models.

While end-to-end models have shown great success on the Automatic Speech Recognition task, performance degrades severely when target sentences are long-form. The previous proposed methods, (partial) overlapping inference are shown to be effective on long-form decoding. For both methods, word error rate (WER) decreases monotonically when over-lapping percentage decreases. Setting aside computational cost, the setup with 50% overlapping during inference can achieve the best performance. However, a lower overlapping percentage has an advantage of fast inference speed. In this paper, we first conduct comprehensive experiments comparing overlapping inference and partial overlapping inference with various configurations. We then propose Voice-Activity-Detection Overlapping Inference to provide a trade-off between WER and computation cost. Results show that the pro-posed method can achieve a 20% relative computation cost reduction on Librispeech and Microsoft Speech Language Translation long-form corpus while maintaining the WER performance when comparing to the best performing over-lapping inference algorithm. We also propose Soft-Match to compensate for similar words misaligned problem.

We present an approach to reduce the performance disparity between geographic regions without degrading performance on the overall user population for ASR.A popular approach is to fine-tune the model with data from regions where the ASR model has a higher word error rate (WER).However, when the ASR model is adapted to get better performance on these high-WER regions, its parameters wander from the previous optimal values, which can lead to worse performance in other regions.In our proposed method, we utilize the elastic weight consolidation (EWC) regularization loss to identify directions in parameters space along which the ASR weights can vary to improve for high-error regions, while still maintaining performance on the speaker population overall.Our results demonstrate that EWC can reduce the word error rate (WER) in the region with highest WER by 3.2% relative while reducing the overall WER by 1.3% relative.We also evaluate the role of language and acoustic models in ASR fairness and propose a clustering algorithm to identify WER disparities based on geographic region.

Contrastive Predictive Coding (CPC) is a representation learning method that maximizes the mutual information between intermediate latent representations and the output of a given model. It can be used to effectively initialize the encoder of an Automatic Speech Recognition (ASR) model. We present a novel modification of CPC called Guided Contrastive Predictive Coding (GCPC). Our proposed method maximizes the mutual information between representations from a prior-knowledge model and the output of the model being pre-trained, allowing prior knowledge injection during pre-training. We validate our method on 3 ASR tasks: German, French and English. Our method outperforms CPC pre-training on all three datasets, reducing the Word Error Rate (WER) by 4.44%, 6.55% and 15.43% relative on the German, French and English (Librispeech) tasks respectively, compared to training from scratch, while CPC pre-training only brings 2.96%, 1.01% and 14.39% relative WER reduction respectively.

Accents mismatching is a critical problem for end-to-end ASR. This paper aims to address this problem by building an accent-robust RNN-T system with domain adversarial training (DAT). We unveil the magic behind DAT and provide, for the first time, a theoretical guarantee that DAT learns accent-invariant representations. We also prove that performing the gradient reversal in DAT is equivalent to minimizing the Jensen-Shannon divergence between domain output distributions. Motivated by the proof of equivalence, we introduce reDAT, a novel technique based on DAT, which relabels data using either unsupervised clustering or soft labels. Experiments on 23K hours of multi-accent data show that DAT achieves competitive results over accent-specific baselines on both native and non-native English accents but up to 13% relative WER reduction on unseen accents; our reDAT yields further improvements over DAT by 3% and 8% relatively on non-native accents of American and British English.

Conventional dynamic language switching enables seamless multilingual interactions by running several monolingual ASR systems in parallel and triggering the appropriate downstream components using a standalone language identification (LID) service. Since this solution is neither scalable nor cost- and memory-efficient, especially for on-device applications, we propose end-to-end, streaming, joint ASR-LID architectures based on the recurrent neural network transducer framework. Two key formulations are explored: (1) joint training using a unified output space for ASR and LID vocabularies, and (2) joint training viewed as multi-task optimization. We also evaluate the benefit of using auxiliary language information obtained on-the-fly from an acoustic LID classifier. Experiments with the English-Hindi language pair show that: (a) multi-task architectures perform better overall, and (b) the best joint architecture surpasses monolingual ASR (6.4–9.2% word error rate reduction) and acoustic LID (53.9–56.1% error rate reduction) baselines while reducing the overall memory footprint by up to 46%.

End-to-end (E2E) automatic speech recognition (ASR) models have recently demonstrated superior performance over the traditional hybrid ASR models. Training an E2E ASR model requires a large amount of data which is not only expensive but may also raise dependency on production data. At the same time, synthetic speech generated by the state-of-the-art text-to-speech (TTS) engines has advanced to near-human naturalness. In this work, we propose to utilize synthetic speech for ASR training (SynthASR) in applications where data is sparse or hard to get for ASR model training. In addition, we apply continual learning with a novel multi-stage training strategy to address catastrophic forgetting, achieved by a mix of weighted multi-style training, data augmentation, encoder freezing, and parameter regularization. In our experiments conducted on in-house datasets for a new application of recognizing medication names, training ASR RNN-T models with synthetic audio via the proposed multi-stage training improved the recognition performance on new application by more than 65% relative, without degradation on existing general applications. Our observations show that SynthASR holds great promise in training the state-of-the-art large-scale E2E ASR models for new applications while reducing the costs and dependency on production data.

Wav2vec-C introduces a novel representation learning technique combining elements from wav2vec 2.0 and VQ-VAE. Our model learns to reproduce quantized representations from partially masked speech encoding using a contrastive loss in a way similar to Wav2vec 2.0. However, the quantization process is regularized by an additional consistency network that learns to reconstruct the input features to the wav2vec 2.0 network from the quantized representations in a way similar to a VQ-VAE model. The proposed self-supervised model is trained on 10k hours of unlabeled data and subsequently used as the speech encoder in a RNN-T ASR model and fine-tuned with 1k hours of labeled data. This work is one of only a few studies of self-supervised learning on speech tasks with a large volume of real far-field labeled data. The Wav2vec-C encoded representations achieves, on average, twice the error reduction over baseline and a higher codebook utilization in comparison to wav2vec 2.0

In this paper, we propose a streaming model to distinguish voice queries intended for a smart-home device from background speech. The proposed model consists of multiple CNN layers with residual connections, followed by a stacked LSTM architecture. The streaming capability is achieved by using unidirectional LSTM layers and a causal mean aggregation layer to form the final utterance-level prediction up to the current frame. In order to avoid redundant computation during online streaming inference, we use a caching mechanism for every convolution operation. Experimental results on a device-directed vs. non device-directed task show that the proposed model yields an equal error rate reduction of 41% compared to our previous best model on this task. Furthermore, we show that the proposed model is able to accurately predict earlier in time compared to the attention-based models.

In this work, we propose a novel and efficient minimum word error rate (MWER) training method for RNN-Transducer (RNN-T).Unlike previous work on this topic, which performs on-the-fly limited-size beam-search decoding and generates alignment scores for expected edit-distance computation, in our proposed method, we re-calculate and sum scores of all the possible alignments for each hypothesis in N-best lists.The hypothesis probability scores and back-propagated gradients are calculated efficiently using the forward-backward algorithm.Moreover, the proposed method allows us to decouple the decoding and training processes, and thus we can perform offline parallel-decoding and MWER training for each subset iteratively.Experimental results show that this proposed semi-on-the-fly method can speed up the on-the-fly method by 6 times and result in a similar WER improvement (3.6%) over a baseline RNN-T model.The proposed MWER training can also effectively reduce high-deletion errors (9.2% WER-reduction) introduced by RNN-T models when EOS is added for endpointer.Further improvement can be achieved if we use a proposed RNN-T rescoring method to re-rank hypotheses and use external RNN-LM to perform additional rescoring.The best system achieves a 5% relative improvement on an English test-set of real far-field recordings and a 11.6% WER reduction on music-domain utterances.

We present a speech data corpus that simulates a "dinner party" scenario taking place in an everyday home environment. The corpus was created by recording multiple groups of four Amazon employee volunteers having a natural conversation in English around a dining table. The participants were recorded by a single-channel close-talk microphone and by five far-field 7-microphone array devices positioned at different locations in the recording room. The dataset contains the audio recordings and human labeled transcripts of a total of 10 sessions with a duration between 15 and 45 minutes. The corpus was created to advance in the field of noise robust and distant speech processing and is intended to serve as a public research and benchmarking data set.

For real-world speech recognition applications, noise robustness is still a challenge. In this work, we adopt the teacher-student (T/S) learning technique using a parallel clean and noisy corpus for improving automatic speech recognition (ASR) performance under multimedia noise. On top of that, we apply a logits selection method which only preserves the k highest values to prevent wrong emphasis of knowledge from the teacher and to reduce bandwidth needed for transferring data. We incorporate up to 8000 hours of untranscribed data for training and present our results on sequence trained models apart from cross entropy trained ones. The best sequence trained student model yields relative word error rate (WER) reductions of approximately 10.1%, 28.7% and 19.6% on our clean, simulated noisy and real test sets respectively comparing to a sequence trained teacher.

We present an end-of-utterance detector for real-time automatic speech recognition in far-field scenarios. The proposed system consists of three components: a long short-term memory (LSTM) neural network trained on acoustic features, an LSTM trained on l-best recognition hypotheses of the automatic speech recognition (ASR) decoder, and a feedforward deep neural network (DNN) combining embeddings derived from both LSTMs with pause duration features from the ASR decoder. At inference time, lower and upper latency (pause duration) bounds act as safeguards. Within the latency bounds, the utterance end-point is triggered as soon as the DNN posterior reaches a tuned threshold. Our experimental evaluation is carried out on real recordings of natural human interactions with voice-controlled far-field devices. We show that the acoustic embeddings are the single most powerful feature and particularly suitable for cross-lingual applications. We furthermore show the benefit of ASR decoder features, especially as a low cost alternative to ASR hypothesis em-beddings.

In this work, we propose a classifier for distinguishing device-directed queries from background speech in the context of interactions with voice assistants.Applications include rejection of false wake-ups or unintended interactions as well as enabling wake-word free followup queries.Consider the example interaction: "Computer, play music", "Computer, reduce the volume".In this interaction, the user needs to repeat the wake-word (Computer) for the second query.To allow for more natural interactions, the device could immediately re-enter listening state after the first query (without wake-word repetition) and accept or reject a potential follow-up as device-directed or background speech.The proposed model consists of two long short-term memory (LSTM) neural networks trained on acoustic features and automatic speech recognition (ASR) 1-best hypotheses, respectively.A feed-forward deep neural network (DNN) is then trained to combine the acoustic and 1-best embeddings, derived from the LSTMs, with features from the ASR decoder.Experimental results show that ASR decoder, acoustic embeddings, and 1-best embeddings yield an equal-error-rate (EER) of 9.3 %, 10.9 % and 20.1 %, respectively.Combination of the features resulted in a 44 % relative improvement and a final EER of 5.2 %.

This article presents a whisper speech detector in the far-field domain. The proposed system consists of a long-short term memory (LSTM) neural network trained on log-filterbank energy (LFBE) acoustic features. This model is trained and evaluated on recordings of human interactions with voice-controlled, far-field devices in whisper and normal phonation modes. We compare multiple inference approaches for utterance-level classification by examining trajectories of the LSTM posteriors. In addition, we engineer a set of features based on the signal characteristics inherent to whisper speech, and evaluate their effectiveness in further separating whisper from normal speech. A benchmarking of these features using multilayer perceptrons (MLP) and LSTMs suggests that the proposed features, in combination with LFBE features, can help us further improve our classifiers. We prove that, with enough data, the LSTM model is indeed as capable of learning whisper characteristics from LFBE features alone compared to a simpler MLP model that uses both LFBE and features engineered for separating whisper and normal speech. In addition, we prove that the LSTM classifiers accuracy can be further improved with the incorporation of the proposed engineered features.

In this paper, we present the elitist particle filter based on evolutionary strategies (EPFES) as an efficient approach to estimate the statistics of a latent state vector capturing the relevant information of a nonlinear system. Similar to classical particle filtering, the EPFES consists of a set of particles and respective weights which represent different realizations of the latent state vector and their likelihood of being the solution of the optimization problem. As main innovation, the EPFES includes an evolutionary elitist-particle selection scheme which combines long-term information with instantaneous sampling from an approximated continuous posterior distribution. In this paper, we propose two advancements of the previously published elitist-particle selection process. Further, the EPFES is shown to be a generalization of the widely-used Gaussian particle filter and thus evaluated with respect to the latter: First, we consider the univariate nonstationary growth model with time-variant latent state variable to evaluate the tracking capabilities of the EPFES for instantaneously calculated particle weights. This is followed by addressing the problem of single-channel nonlinear acoustic echo cancellation as a challenging benchmark task for identifying an unknown system of large search space: the nonlinear acoustic echo path is modeled by a cascade of a parameterized preprocessor (to model the loudspeaker signal distortions) and a linear FIR filter (to model the sound wave propagation and the microphone). By using long-term information, we highlight the efficacy of the well-generalizing EPFES in estimating the preprocessor parameters for a simulated scenario and a real smartphone recording. Finally, we illustrate similarities between the EPFES and evolutionary algorithms to outline future improvements by fusing the achievements of both fields of research.

Uncertainty decoding combines a probabilistic feature description with the acoustic model of a speech recognition system. For DNN-HMM hybrid systems, this can be realized by averaging the DNN outputs produced by a finite set of feature samples (drawn from an estimated probability distribution). In this article, we employ this sampling approach in combination with a multi-microphone speech enhancement system. We propose a new strategy for generating feature samples from multichannel signals, based on modeling the spatial coherence estimates between different microphone pairs as realizations of a latent random variable. From each coherence estimate, a spectral enhancement gain is computed and an enhanced feature vector is obtained, thus producing a finite set of feature samples, of which we average the respective DNN outputs. In the experimental part, this new uncertainty decoding strategy is shown to consistently improve the recognition accuracy of a DNN-HMM hybrid system for the 8-channel REVERB Challenge task.

We propose a spatial diffuseness feature for deep neural network (DNN)-based automatic speech recognition to improve recognition accuracy in reverberant and noisy environments. The feature is computed in real-time from multiple microphone signals without requiring knowledge or estimation of the direction of arrival, and represents the relative amount of diffuse noise in each time and frequency bin. It is shown that using the diffuseness feature as an additional input to a DNN-based acoustic model leads to a reduced word error rate for the REVERB challenge corpus, both compared to logmelspec features extracted from noisy signals, and features enhanced by spectral subtraction.

In this letter, we derive a new stepsize adaptation for the normalized least mean square algorithm (NLMS) by describing the task of linear acoustic echo cancellation from a Bayesian network perspective. Similar to the well-known Kalman filter equations, we model the acoustic wave propagation from the loudspeaker to the microphone by a latent state vector and define a linear observation equation (to model the relation between the state vector and the observation) as well as a linear process equation (to model the temporal progress of the state vector). Based on additional assumptions on the statistics of the random variables in observation and process equation, we apply the expectation-maximization (EM) algorithm to derive an NLMS-like filter adaptation. By exploiting the conditional independence rules for Bayesian networks, we reveal that the resulting EM-NLMS algorithm has a stepsize update equivalent to the optimal-stepsize calculation proposed by Yamamoto and Kitayama in 1982, which has been adopted in many textbooks. As main difference, the instantaneous stepsize value is estimated in the M step of the EM algorithm (instead of being approximated by artificially extending the acoustic echo path). The EM-NLMS algorithm is experimentally verified for synthesized scenarios with both, white noise and male speech as input signal.

In this article, we introduce a novel approach for nonlinear acoustic echo cancellation based on a combination of particle filtering and evolutionary strategies. The nonlinear echo path is modeled as a state vector with non-Gaussian probability distribution and the relation to the observed signals and near-end interferences are captured by nonlinear functions. To estimate the probability distribution of the state vector and the model parameters, we apply the numerical sampling method of particle filtering, where each set of particles represents different realizations of the nonlinear echo path. While the classical particle-filter approach is unsuitable for system identification with large search spaces, we introduce a modified particle filter to select elitist particles based on long-term fitness measures and to create new particles based on the approximated probability distribution of the state vector. The validity of the novel approach is experimentally verified with real recordings for a nonlinear echo path stemming from a commercial smartphone.

The generic REMOS (REverberation MOdeling for robust Speech recognition) concept is extended in this contribution to cope with additional noise components. REMOS originally embeds an explicit reverberation model into a hiddenMarkov model (HMM) leading to a relaxed conditional independence assumption for the observed feature vectors. During recognition, a nonlinear optimization problem is to be solved in order to adapt the HMMs' output probability density functions to the current reverberation conditions. The extension for additional noise components necessitates a modified numerical solver for the nonlinear optimization problem. We propose an approximation scheme based on continuous piecewise linear regression. Connected-digit recognition experiments demonstrate the potential of REMOS in reverberant and noisy environments. They furthermore reveal that the benefit of an explicit reverberation model, overcoming the conditional independence assumption, increases with increasing signal-to-noise-ratios.

In this paper, we study the effect of the design parameters of a single-channel reverberation suppression algorithm on reverberation-robust speech recognition. At the same time, reverberation compensation at the speech recognizer is investigated. The analysis reveals that it is highly beneficial to attenuate only the reverberation tail after approximately 50 ms while coping with the early reflections and residual late-reverberation by training the recognizer on moderately reverberant data. It will be shown that the overall system at its optimum configuration yields a very promising recognition performance even in strongly reverberant environments. Since the reverberation suppression algorithm is evidenced to significantly reduce the dependency on the training data, it allows for a very efficient training of acoustic models that are suitable for a wide range of reverberation conditions. Finally, experiments with an “ideal” reverberation suppression algorithm are carried out to cross-check the inferred guidelines.

A novel frame-wise model adaptation approach for reverberation robust distant-talking speech recognition is proposed. It adjusts the means of static cepstral features to capture the statistics of reverber ant feature vector sequences obtained from distant-talking speech recordings. The means of the HMMs are adapted during decoding using a state-dependent estimate of the late reverberation determined by joint use of a feature-domain reverberation model and optimum partial state sequences. Since the parameters of the HMMs and the reverberation model can be estimated completely independently, the approach is very flexible with respect to changing acoustic environments. Due to the frame-wise model adaptation, some of the HMM limitations are relieved, and recognition results surpassing that of matched reverberant training are obtained at the cost of a moderately increased decoding complexity.

The REMOS (REverberation MOdeling for Speech recognition) concept for reverberation-robust distant-talking speech recognition, introduced in for melspectral features, is extended in this contribution to logarithmic melspectral (logmelspec) features. Based on a combined acoustic model consisting of a hidden Markov model network and a reverberation model, REMOS determines clean-speech and reverberation estimates during recognition by an inner optimization operation. A reformulation of this inner optimization problem for logmelspec features, allowing an efficient solution by nonlinear optimization algorithms, is derived in this paper so that an efficient implementation of REMOS for logmelspec features becomes possible. Connected digit recognition experiments show that the proposed REMOS implementation significantly outperforms reverberantly-trained HMMs in highly reverberant environments.

The REMOS (REverberation MOdeling for Speech recognition) concept for reverberation-robust distant-talking speech recognition, introduced in “Distant-talking continuous speech recognition based on a novel reverberation model in the feature domain” (A. Sehr , in Proc. Interspeech, 2006, pp. 769-772) for melspectral features, is extended to logarithmic melspectral (logmelspec) features in this contribution. Thus, the favorable properties of REMOS, including its high flexibility with respect to changing reverberation conditions, become available in the more competitive logmelspec domain. Based on a combined acoustic model consisting of a hidden Markov model (HMM) network and a reverberation model (RM), REMOS determines clean-speech and reverberation estimates during recognition. Therefore, in each iteration of a modified Viterbi algorithm, an inner optimization operation maximizes the joint density of the current HMM output and the RM output subject to the constraint that their combination is equal to the current reverberant observation. Since the combination operation in the logmelspec domain is nonlinear, numerical methods appear necessary for solving the constrained inner optimization problem. A novel reformulation of the constraint, which allows for an efficient solution by nonlinear optimization algorithms, is derived in this paper so that a practicable implementation of REMOS for logmelspec features becomes possible. An in-depth analysis of this REMOS implementation investigates the statistical properties of its reverberation estimates and thus derives possibilities for further improving the performance of REMOS. Connected digit recognition experiments show that the proposed REMOS version in the logmelspec domain significantly outperforms the melspec version. While the proposed RMs with parameters estimated by straightforward training for a given room are robust to a mismatch of the speaker-microphone distance, their performance significantly decreases if they are used in a room with substantially different conditions. However, by training multi-style RMs with data from several rooms, good performance can be achieved across different rooms.