End-to-end automatic speech recognition (ASR) models have seen revolutionary quality gains with the recent development of large-scale universal speech models (USM). However, deploying these massive USMs is extremely expensive due to the enormous memory usage and computational cost. Therefore, model compression is an important research topic to fit USM-based ASR under budget in real-world scenarios. In this study, we propose a USM fine-tuning approach for ASR, with a low-bit quantization and N:M structured sparsity aware paradigm on the model weights, reducing the model complexity from parameter precision and matrix topology perspectives. We conducted extensive experiments with a 2-billion parameter USM on a large-scale voice search dataset to evaluate our proposed method. A series of ablation studies validate the effectiveness of up to int4 quantization and 2:4 sparsity. However, a single compression technique fails to recover the performance well under extreme setups including int2 quantization and 1:4 sparsity. By contrast, our proposed method can compress the model to have 9.4% of the size, at the cost of only 7.3% relative word error rate (WER) regressions. We also provided in-depth analyses on the results and discussions on the limitations and potential solutions, which would be valuable for future studies.

The accuracy of end-to-end (E2E) automatic speech recognition (ASR) models continues to improve as they are scaled to larger sizes, with some now reaching billions of parameters. Widespread deployment and adoption of these models, however, requires computationally efficient strategies for decoding. In the present work, we study one such strategy: applying multiple frame reduction layers in the encoder to compress encoder outputs into a small number of output frames. While similar techniques have been investigated in previous work, we achieve dramatically more reduction than has previously been demonstrated through the use of multiple funnel reduction layers. Through ablations, we study the impact of various architectural choices in the encoder to identify the most effective strategies. We demonstrate that we can generate one encoder output frame for every 2.56 sec of input speech, without significantly affecting word error rate on a large-scale voice search task, while improving encoder and decoder latencies by 48% and 92% respectively, relative to a strong but computationally expensive baseline.

Weiran Wang, Rohit Prabhavalkar, Haozhe Shan, Zhong Meng, Dongseong Hwang, Qiujia Li, Khe Chai Sim, Bo Li, James Qin, Xingyu Cai, Adam Stooke, Chengjian Zheng, Yanzhang He, Tara Sainath, Pedro Moreno Mengibar. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

With the rapid increase in the size of neural networks, model compression has become an important area of research. Quantization is an effective technique for decreasing the size, memory access, and compute load of large models. In this paper, we first benchmark the impact of techniques such as straight through estimator, pseudo-quantization noise (PQN), learnable scale parameter, clipping, etc. on 4-bit seq 2 seq models across a suite of speech recognition datasets ranging from 1,000 hours to 1 million hours, as well as one machine translation dataset to illustrate its applicability outside of speech. It is commonly believed that having dedicated learnable scale parameters per quantization group is critical to model accuracy. Instead, we propose to construct the scale as a Lpnorm of a few of the largest outliers within the quantization group and regularizing that norm within the end-to-end optimization. This outperforms popular learnable scale and clipping methods without the need to introduce extra parameters. PQN-QAT shows a larger improvement under the proposed method, and it opens up the potential to exploit some of its other benefits: 1) training a single model that performs well in mixed precision and 2) improved generalization on long form speech.

In this paper, we explore various frame rate reduction schemes on the two-pass cascaded encoder model to improve its efficiency without scarifying the transcription quality. We conduct extensive studies on frame rate reduction strategies, left and right context window length, trade-offs in quality, latency, computation and power consumption, and performance in short-and long-form datasets. With the proposed schemes, we can lower the 2nd pass frame rate to $120 \mathrm{~ms}$, half of the 1st pass’s. This achieves $20 \%$ RTF reduction / $13 \%$ power saving / $19 \%$ lower final latency, without impact on the word-error-rate nor partial results’ latency. If allowing partial latency increase, we can further reduce the frame rate to $180 \mathrm{~ms}$ or even $240 \mathrm{~ms}$ from the 1st pass, and obtain $45 \%$ RTF / 35% power savings, with a similar or even better (on the short-form testset) recognition accuracy.

With the growing need for large models in speech recognition, quantization has become a valuable technique to reduce their compute and memory transfer costs. Quantized models tends to show worse accuray compared to their float counterparts, due to the noise introduced from quantization. One pathological case is when the noise caused by a group of weights are constructively combined. In contrast to recent advances in post-training quantization, this paper proposes to discourage that constructive combination during the quantization aware training process. This is accomplished by regularizing three different functions of the layer-wise input feature correlation matrix. We show that these methods can improve the 2-bit performance of the Conformer model on LibriSpeech. Additionally, they serve as an informative measure for the ease of quantization in the different types of layers in the Conformer, which we show to help aid selecting layer precision for mixed precision models.

Continued improvements in machine learning techniques offer exciting new opportunities through the use of larger models and larger training datasets. However, there is a growing need to offer these new capabilities on-board low-powered devices such as smart-phones, wearables and other embedded environments where only low memory is available. Towards this, we consider methods to reduce the model size of Conformer-based speech recognition models which typically require models with greater than 100M parameters down to just 5M parameters while minimizing impact on model quality. Such a model allows us to achieve always-on ambient speech recognition on edge devices with low-memory neural processors. We propose model weight reuse at different levels within our model architecture: (i) repeating full conformer block layers, (ii) sharing specific conformer modules across layers, (iii) sharing sub-components per conformer module, and (iv) sharing decomposed sub-component weights after low-rank decomposition. By sharing weights at different levels of our model, we can retain the full model in-memory while increasing the number of virtual trans-formations applied to the input. Through a series of ablation studies and evaluations, we find that with weight sharing and a low-rank architecture, we can achieve a WER of 2.84 and 2.94 for Librispeech dev-clean and test-clean respectively with a 5M parameter model.

We explore unifying a neural segmenter with two-pass cascaded encoder ASR into a single model. A key challenge is allowing the segmenter (which runs in real-time, synchronously with the decoder) to finalize the non-causal 2nd pass (which runs 900 ms behind real-time) without introducing user-perceived latency or deletion errors during inference. We propose a design where the neural segmenter is integrated with the causal 1st pass decoder to emit a end-of-segment (EOS) signal in real-time. The EOS signal is then used to finalize the non-causal 2nd pass. We experiment with different ways to finalize the 2nd pass, and find that a dummy frame injection strategy allows for simultaneous high quality 2nd pass results and low finalization latency. On a real-world long-form captioning task (YouTube), we achieve 2.4% relative WER and 140 ms EOS latency gains over a baseline VAD-based segmenter with the same cascaded encoder.

Recently, Feature-wise Linear Modulation (FiLM) has been shown to outperform other approaches to incorporate speaker embedding into speech separation and VoiceFilter models. We propose an improved method of incorporating such embeddings into a Voice- Filter frontend for automatic speech recognition (ASR) and text- independent speaker verification (TI-SV). We extend the widely- used Conformer architecture to construct a FiLM Block with additional feature processing before and after the FiLM layers. Apart from its application to single-user VoiceFilter, we show that our system can be easily extended to multi-user VoiceFilter models via element-wise max pooling of the speaker embeddings in a projected space. The final architecture, which we call Conditional Conformer, tightly integrates the speaker embeddings into a Conformer backbone. We improve TI-SV equal error rates by as much as 56% over prior multi-user VoiceFilter models, and our element-wise max pooling reduces relative WER compared to an attention mechanism by as much as 10%.

We propose a multi-output joint network architecture for RNN-T transducer, for multi-task modeling of ASR and auxiliary tasks that rely on ASR outputs. Each output of the joint network predicts tar-get labels with disjoint vocabularies for each task, while sharing the same audio features by the encoder and language model features by the prediction network. Each task is trained with an RNN-T loss that marginalizes over all possible paths, and we allow multiple tasks to share the blank logit so that they are synchronized. We demonstrate our method on two auxiliary tasks, namely capitalization and pause prediction, and discuss different considerations for modeling and inference procedures. For capitalization, we successfully distill capitalization labels from a standalone text normalization model, and achieve competitive Uppercase Error Rate (UER) while offering streaming capability and improved inference efficiency. In addition, our model has similar capitalization accuracy compared to a mixed-case ASR model, but obtains improved WERs if integrated with external language models. For pause prediction, we achieve the same performance as the previous two-step approach while providing a simpler training recipe without affecting ASR accuracy.

Self- and semi-supervised learning methods have been actively investigated to reduce labeled training data or enhance model performance. However, these approaches mostly focus on in-domain performance for public datasets. In this study, we utilize the combination of self- and semi-supervised learning methods to solve unseen domain adaptation problems in a large-scale production setting for online ASR model. This approach demonstrates that using the source domain data with a small fraction of the target domain data (3%) can recover the performance gap compared to a full data baseline: 13.5% relative WER improvement for target domain data.

As end-to-end automatic speech recognition (ASR) models reach promising performance, various downstream tasks rely on good confidence estimators for these systems. Recent research has shown that model-based confidence estimators have a significant advantage over using the output softmax probabilities. If the input data to the speech recogniser is from mismatched acoustic and linguistic conditions, the ASR performance and the corresponding confidence estimators may exhibit severe degradation. Since confidence models are often trained on the same in-domain data as the ASR, generalising to out-of-domain (OOD) scenarios is challenging. By keeping the ASR model untouched, this paper proposes two approaches to improve the model-based confidence estimators on OOD data: using pseudo transcriptions and an additional OOD language model. With an ASR model trained on LibriSpeech, experiments show that the proposed methods can greatly improve the confidence metrics on TED-LIUM and Switchboard datasets while preserving in-domain performance. Furthermore, the improved confidence estimators are better calibrated on OOD data and can provide a much more reliable criterion for data selection.

In this paper, we explore reducing computational latency of the 2-pass cascaded encoder model [1]. Specifically, we experiment with reducing the size of the causal 1st-pass and adding capacity to the non-causal 2nd-pass, such that the overall latency can be reduced without loss of quality. In addition, we explore using a confidence model for deciding to stop 2nd-pass recognition if we are confident in the 1st-pass hypothesis. Overall, we are able to reduce latency by a factor of 1.7X, compared to the baseline cascaded encoder from [1]. Secondly, with the added capacity in the non-causal 2nd-pass, we find that we can improve WER by up to 7% relative using wav2vec and minimum word-error-rate (MWER) training.

Reducing the latency and model size has always been a significant research problem for live Automatic Speech Recognition (ASR) application scenarios. Along this direction, model quantization has become an increasingly popular approach to compress neural networks and reduce computation cost. Most of the existing practical ASR systems apply post-training 8-bit quantization. To achieve a higher compression rate without introducing additional performance regression, in this study, we propose to develop 4-bit ASR models with native quantization aware training, which leverages native integer operations to effectively optimize both training and inference. We conducted two experiments on state-of-the-art Conformer-based ASR models to evaluate our proposed quantization technique. First, we explored the impact of different precisions for both weight and activation quantization on the LibriSpeech dataset, and obtained a lossless 4-bit Conformer model with 7.7x size reduction compared to the float32 model. Following this, we for the first time investigated and revealed the viability of 4-bit quantization on a practical ASR system that is trained with large-scale datasets, and produced a lossless Conformer ASR model with mixed 4-bit and 8-bit weights that has 5x size reduction compared to the float32 model.

Personalization of on-device speech recognition (ASR) has seen explosive growth in recent years, largely due to the increasing popularity of personal assistant features on mobile devices and smart home speakers. In this work, we present Personal VAD 2.0, a personalized voice activity detector that detects the voice activity of a target speaker, as part of a streaming on-device ASR system. Although previous proof-of-concept studies have validated the effectiveness of Personal VAD, there are still several critical challenges to address before this model can be used in production: first, the quality must be satisfactory in both enrollment and enrollment-less scenarios; second, it should operate in a streaming fashion; and finally, the model size should be small enough to fit a limited latency and CPU/Memory budget. To meet the multi-faceted requirements, we propose a series of novel designs: 1) advanced speaker embedding modulation methods; 2) a new training paradigm to generalize to enrollment-less conditions; 3) architecture and runtime optimizations for latency and resource restrictions. Extensive experiments on a realistic speech recognition system demonstrated the state-of-the-art performance of our proposed method.

In this paper, we propose a dynamic cascaded encoder Automatic Speech Recognition (ASR) model, which unifies models for different deployment scenarios. Moreover, the model can significantly reduce model size and power consumption without loss of quality. Namely, with the dynamic cascaded encoder model, we explore three techniques to maximally boost the performance of each model size: 1) Use separate decoders for each sub-model while sharing the encoders; 2) Use funnel-pooling to improve the encoder efficiency; 3) Balance the size of causal and non-causal encoders to improve quality and fit deployment constraints. Overall, the proposed large-medium model has 30% smaller size and reduces power consumption by 33%, compared to the baseline cascaded encoder model. The triple-size model that unifies the large, medium, and small models achieves 37% total size reduction with minimal quality loss, while substantially reducing the engineering efforts of having separate models.

Text-only and semi-supervised training based on audio-only data has gained popularity recently due to the wide availability of unlabeled text and speech data.In this work, we propose incorporating text-only and semi-supervised training into an attention-based deliberation model.By incorporating textonly data in training a bidirectional encoder representation from transformer (BERT) for the deliberation text encoder, and large-scale text-to-speech and audio-only utterances using joint acoustic and text decoder (JATD) and semi-supervised training, we achieved 4%-12% WER reduction for various tasks compared to the baseline deliberation.Compared to a state-of-theart language model (LM) rescoring method, the deliberation model reduces the Google Voice Search WER by 11% relative.We show that the deliberation model also achieves a positive human side-by-side evaluation compared to the state-of-the-art LM rescorer with reasonable endpointer latencies.

On-device end-to-end (E2E) models have shown improvements over a conventional model on English Voice Search tasks in both quality and latency.E2E models have also shown promising results for multilingual automatic speech recognition (ASR).In this paper, we extend our previous capacity solution to streaming applications and present a streaming multilingual E2E ASR system that runs fully on device with comparable quality and latency to individual monolingual models.To achieve that, we propose an Encoder Endpointer model and an End-of-Utterance (EOU) Joint Layer for a better quality and latency trade-off.Our system is built in a language agnostic manner allowing it to natively support intersentential code switching in real time.To address the feasibility concerns on large models, we conducted on-device profiling and replaced the time consuming LSTM decoder with the recently developed Embedding decoder.With these changes, we managed to run such a system on a mobile device in less than real time.

Language models (LMs) significantly improve the recognition accuracy of end-to-end (E2E) models on words rarely seen during training, when used in either the shallow fusion or the rescoring setups. In this work, we introduce LMs in the learning of hybrid autoregressive transducer (HAT) models in the discriminative training framework, to mitigate the training versus inference gap regarding the use of LMs. For the shallow fusion setup, we use LMs during both hypotheses generation and loss computation, and the LM-aware MWER-trained model achieves 10\% relative improvement over the model trained with standard MWER on voice search test sets containing rare words. For the rescoring setup, we learn a small neural module to generate per-token fusion weights in a data-dependent manner. This model achieves the same rescoring WER as regular MWER-trained model, but without the need for sweeping fusion weights.

Automatic speech recognition (ASR) systems typically rely on an external endpointer (EP) model to identify speech boundaries. In this work, we propose a method to jointly train the ASR and EP tasks in a single end-to-end (E2E) multitask model, improving EP quality by optionally leveraging information from the ASR audio encoder. We introduce a “switch” connection, which trains the EP to consume either the audio frames directly or low-level latent representations from the ASR model. This results in a single E2E model that can be used during inference to perform frame filtering at low cost, and also make high quality end-of-query (EOQ) predictions based on ongoing ASR computation. We present results on a voice search test set showing that, compared to separate single-task models, this approach reduces median endpoint latency by 120 ms (30.8% reduction), and 90th percentile latency by 170 ms (23.0% reduction), without regressing word error rate. For continuous recognition, WER improves by 10.6% (relative).

Incremental speech recognizers start displaying results while the users are still speaking. These partial results are beneficial to users who like the responsiveness of the system. However, as new partial results come in, words that were previously displayed can change or disappear. The results appear unstable and this unwanted phenomenon is called flickering. Typical remediation approaches can increase latency and reduce the quality of the partials results, but little work has been done to measure these effects. We first introduce two new metrics that allow us to measure the quality and latency of the partials. We propose the new, lightweight approach of reranking the partial results in favor of a more stable prefix without changing the beam search. This allows us to reduce flickering without impacting the final result. We show that we can roughly halve the amount of flickering with negligible impact on the quality and latency of the partial results.

Confidence estimation for automatic speech recognition (ASR) is important for many downstream tasks. Recently, neural confidence estimation models (CEMs) have been shown to produce accurate confidence scores for predicting word-level errors. These models are built on top of an end-to-end (E2E) ASR and the acoustic embeddings are part of the input features. However, practical E2E ASR systems often incorporate contextual information in the decoder to improve rare word recognition. The CEM is not aware of this and underestimates the confidence of the rare words that have been corrected by the context. In this paper, we propose a context-aware CEM by incorporating context into the encoder using a neural associative memory (NAM) model. It uses attention to detect for presence of the biasing phrases and modify the encoder features. Experiments show that the proposed context-aware CEM using NAM augmented training can improve the AUC-ROC for word error prediction from 0.837 to 0.892.

This work introduces cross-attention conformer, an attention-based architecture for context modeling in speech enhancement. Given that the context information can often be sequential, and of different length as the audio that is to be enhanced, we make use of cross-attention to summarize and merge contextual information with input features. Building upon the recently proposed conformer model that uses self attention layers as building blocks, the proposed cross-attention conformer can be used to build deep contextual models. As a concrete example, we show how noise context, i.e., short noise-only audio segment preceding an utterance, can be used to build a speech enhancement feature frontend using cross-attention conformer layers for improving noise robustness of automatic speech recognition.

In this paper, we propose a solution to allow speaker conditioned speech models, such as VoiceFilter-Lite, to support an arbitrary number of enrolled users in a single pass. This is achieved via an attention mechanism on multiple speaker embeddings to compute a single attentive embedding, which is then used as a side input to the model. We implemented multi-user VoiceFilter-Lite and evaluated it for three tasks: (1) a streaming automatic speech recognition (ASR) task; (2) a text-independent speaker verification task; and (3) a personalized keyphrase detection task, where ASR has to detect keyphrases from multiple enrolled users in a noisy environment. Our experiments show that, with up to four enrolled users, multi-user VoiceFilter-Lite is able to significantly reduce speech recognition and speaker verification errors when there is overlapping speech, without affecting performance under other acoustic conditions. This attentive speaker embedding approach can also be easily applied to other speaker-conditioned models such as personal voice activity detection (VAD) and personalized ASR.

End-to-end (E2E) models have shown to outperform state-of-the-art conventional models for streaming speech recognition [1] across many dimensions, including quality (as measured by word error rate (WER)) and endpointer latency [2]. However, the model still tends to delay the predictions towards the end and thus has much higher partial latency compared to a conventional ASR model. To address this issue, we look at encouraging the E2E model to emit words early, through an algorithm called FastEmit [3]. Naturally, improving on latency results in a quality degradation. To address this, we explore replacing the LSTM layers in the encoder of our E2E model with Conformer layers [4], which has shown good improvements for ASR. Secondly, we also explore running a 2nd-pass beam search to improve quality. In order to ensure the 2nd-pass completes quickly, we explore non-causal Conformer layers that feed into the same 1st-pass RNN-T decoder, an algorithm called Cascaded Encoders [5]. Overall, the Conformer RNN-T with Cascaded Encoders offers a better quality and latency tradeoff for streaming ASR.

For various speech-related tasks, confidence scores from a speech recogniser are a useful measure to assess the quality of transcriptions. In traditional hidden Markov model-based automatic speech recognition (ASR) systems, confidence scores can be reliably obtained from word posteriors in decoding lattices. However, for an ASR system with an auto-regressive decoder, such as an attention-based sequence-to-sequence model, computing word posteriors is difficult. An obvious alternative is to use the decoder softmax probability as the model confidence. In this paper, we first examine how some commonly used regularisation methods influence the softmax-based confidence scores and study the overconfident behaviour of end-to-end models. Then we propose a lightweight and effective approach named confidence estimation module (CEM) on top of an existing end-to-end ASR model. Experiments on LibriSpeech show that CEM can mitigate the overconfidence problem and can produce more reliable confidence scores with and without shallow fusion of a language model. Further analysis shows that CEM generalises well to speech from a moderately mismatched domain and can potentially improve downstream tasks such as semi-supervised learning.

End-to-end models that condition the output sequence on all previously predicted labels have emerged as popular alternatives to conventional systems for automatic speech recognition (ASR). Since distinct label histories correspond to distinct models states, such models are decoded using an approximate beam-search which produces a tree of hypotheses.In this work, we study the influence of the amount of label context on the model’s accuracy, and its impact on the efficiency of the decoding process. We find that we can limit the context of the recurrent neural network transducer (RNN-T) during training to just four previous word-piece labels, without degrading word error rate (WER) relative to the full-context baseline. Limiting context also provides opportunities to improve decoding efficiency by removing redundant paths from the active beam, and instead retaining them in the final lattice. This path-merging scheme can also be applied when decoding the baseline full-context model through an approximation. Overall, we find that the proposed path-merging scheme is extremely effective, allowing us to improve oracle WERs by up to 36% over the baseline, while simultaneously reducing the number of model evaluations by up to 5.3% without any degradation in WER, or up to 15.7% when lattice rescoring is applied.

We study the problem of word-level confidence estimation in subword-based end-to-end (E2E) models for automatic speech recognition (ASR). Although prior works have proposed training auxiliary confidence models for ASR systems, they do not extend naturally to systems that operate on word-pieces (WP) as their vocabulary. In particular, ground truth WP correctness labels are needed for training confidence models, but the non-unique tokenization from word to WP causes inaccurate labels to be generated. This paper proposes and studies two confidence models of increasing complexity to solve this problem. The final model uses self-attention to directly learn word-level confidence without needing subword tokenization, and exploits full context features from multiple hypotheses to improve confidence accuracy. Experiments on Voice Search and long-tail test sets show standard metrics (e.g., NCE, AUC, RMSE) improving substantially. The proposed confidence module also enables a model selection approach to combine an on-device E2E model with a hybrid model on the server to address the rare word recognition problem for the E2E model.

Streaming automatic speech recognition (ASR) aims to emit each hypothesized word as quickly and accurately as possible. However, emitting fast without degrading quality, as measured by word error rate (WER), is highly challenging. Existing approaches including Early and Late Penalties [1] and Constrained Alignments [2], [3] penalize emission delay by manipulating per-token or per-frame probability prediction in sequence transducer models [4]. While being successful in reducing delay, these approaches suffer from significant accuracy regression and also require additional word alignment information from an existing model. In this work, we propose a sequence-level emission regularization method, named FastEmit, that applies latency regularization directly on per-sequence probability in training transducer models, and does not require any alignment. We demonstrate that FastEmit is more suitable to the sequence-level optimization of transducer models [4] for streaming ASR by applying it on various end-to-end streaming ASR networks including RNN-Transducer [5], Transformer-Transducer [6], [7], ConvNet-Transducer [8] and Conformer-Transducer [9]. We achieve 150 ~ 300ms latency reduction with significantly better accuracy over previous techniques on a Voice Search test set. FastEmit also improves streaming ASR accuracy from 4.4%/8.9% to 3.1%/7.5% WER, meanwhile reduces 90th percentile latency from 210ms to only 30ms on LibriSpeech.

Previous works on the Recurrent Neural Network-Transducer (RNN-T) models have shown that, under some conditions, it is possible to simplify its prediction network with little or no loss in recognition accuracy (arXiv:2003.07705 [eess.AS], [2], arXiv:2012.06749 [cs.CL]). This is done by limiting the context size of previous labels and/or using a simpler architecture for its layers instead of LSTMs. The benefits of such changes include reduction in model size, faster inference and power savings, which are all useful for on-device applications. In this work, we study ways to make the RNN-T decoder (prediction network + joint network) smaller and faster without degradation in recognition performance. Our prediction network performs a simple weighted averaging of the input embeddings, and shares its embedding matrix weights with the joint network's output layer (a.k.a. weight tying, commonly used in language modeling arXiv:1611.01462 [cs.LG]). This simple design, when used in conjunction with additional Edit-based Minimum Bayes Risk (EMBR) training, reduces the RNN-T Decoder from 23M parameters to just 2M, without affecting word-error rate (WER).

Confidence scores are very useful for downstream applications of automatic speech recognition (ASR) systems. Recent works have proposed using neural networks to learn word or utterance confidence scores for end-to-end ASR. In those studies, word confidence by itself does not model deletions, and utterance confidence does not take advantage of word-level training signals. This paper proposes to jointly learn word confidence, word deletion, and utterance confidence. Empirical results show that multi-task learning with all three objectives improves confidence metrics (NCE, AUC, RMSE) without the need for increasing the model size of the confidence estimation module. Using the utterance-level confidence for rescoring also decreases the word error rates on Google's Voice Search and Long-tail Maps datasets by 3-5% relative, without needing a dedicated neural rescorer.

In this paper, we introduce a streaming keyphrase detection system that can be easily customized to accurately detect any phrase composed of words from a large vocabulary. The system is implemented with an end-to-end trained automatic speech recognition (ASR) model and a text-independent speaker verification model. To address the challenge of detecting these keyphrases under various noisy conditions, a speaker separation model is added to the feature frontend of the speaker verification model, and an adaptive noise cancellation (ANC) algorithm is included to exploit cross-microphone noise coherence. Our experiments show that the text-independent speaker verification model largely reduces the false triggering rate of the keyphrase detection, while the speaker separation model and adaptive noise cancellation largely reduce false rejections.

End-to-end (E2E) models fold the acoustic, pronunciation and language models of a conventional speech recognition model into one neural network with a much smaller number of parameters than a conventional ASR system, thus making it suitable for on-device applications. For example, recurrent neural network transducer (RNN-T) as a streaming E2E model has shown promising potential for on-device ASR [1]. For such applications, quality and latency are two critical factors. We propose to reduce E2E model's latency by extending the RNN-T endpointer (RNN-T EP) model [2] with additional early and late penalties. By further applying the minimum word error rate (MWER) training technique [3], we achieved 8.0% relative word error rate (WER) reduction and 130ms 90-percentile latency reduction over [2] on a Voice Search test set. We also experimented with a second-pass Listen, Attend and Spell (LAS) rescorer [4]. Although it did not directly improve the first pass latency, the large WER reduction provides extra room to trade WER for latency. RNN-T EP+LAS, together with MWER training brings in 18.7% relative WER reduction and 160ms 90-percentile latency reductions compared to the original proposed RNN-T EP [2] model.

Thus far, end-to-end (E2E) models have not been shown to outperform state-of-the-art conventional models with respect to both quality, i.e., word error rate (WER), and latency, i.e., the time the hypothesis is finalized after the user stops speaking. In this paper, we develop a first-pass Recurrent Neural Network Transducer (RNN-T) model and a second-pass Listen, Attend, Spell (LAS) rescorer that surpasses a conventional model in both quality and latency. On the quality side, we incorporate a large number of utterances across varied domains [1] to increase acoustic diversity and the vocabulary seen by the model. We also train with accented English speech to make the model more robust to different pronunciations. In addition, given the increased amount of training data, we explore a varied learning rate schedule. On the latency front, we explore using the end-of-sentence decision emitted by the RNN-T model to close the microphone, and also introduce various optimizations to improve the speed of LAS rescoring. Overall, we find that RNN-T+LAS offers a better WER and latency tradeoff compared to a conventional model. For example, for the same latency, RNN-T+LAS obtains a 8% relative improvement in WER, while being more than 400-times smaller in model size.

Recently, we introduced a two-pass on-device end-to-end (E2E) speech recognition model, which runs RNN-T in the first-pass and then rescores/redecodes the result using a noncausal Listen, Attend and Spell (LAS) decoder. This on-device model obtained similar performance to a state-of-the-art conventional model. However, like many E2E models, it suffers from being trained only on supervised audio-text pairs and thus performs poorly on rare words compared to a conventional model which incorporates a language model trained on a much larger text corpus. In this work, we introduce a joint acoustic and text decoder (JATD) into the LAS decoder, which makes it possible to incorporate a much larger text corpus into training. We find that the JATD model obtains in a 3-10% relative improvement in WER compared to a LAS decoder trained only on supervised audio-text pairs across a variety of proper noun test sets.

Recent advances of end-to-end models have outperformed conventional models through employing a two-pass model. The two-pass model provides better speed-quality trade-offs for on-device speech recognition, where a 1st-pass model generates hypotheses in a streaming fashion, and a 2nd-pass model re-scores the hypotheses with full audio sequence context. The 2nd-pass model plays a key role in the quality improvement of the end-to-end model to surpass the conventional model. One main challenge of the two-pass model is the computation latency introduced by the 2nd-pass model. Specifically, the original design of the two-pass model uses LSTMs for the 2nd-pass model, which are subject to long latency as they are constrained by the recurrent nature and have to run inference sequentially. In this work we explore replacing the LSTM layers in the 2nd-pass rescorer with Transformer layers, which can process the entire hypothesis sequences in parallel and can therefore utilize the on-device computation resources more efficiently. Compared with an LSTM-based baseline, our proposed Transformer rescorer achieves more than 50% latency reduction with quality improvement.

The demand for fast and accurate incremental speech recognition increases as the applications of automatic speech recognition (ASR) proliferate.Incremental speech recognizers output chunks of partially recognized words while the user is still talking.Partial results can be revised before the ASR finalizes its hypothesis, causing instability issues.We analyze the quality and stability of on-device streaming end-to-end (E2E) ASR models.We first introduce a novel set of metrics that quantify the instability at word and segment levels.We study the impact of several model training techniques that improve E2E model qualities but degrade model stability.We categorize the causes of instability and explore various solutions to mitigate them in a streaming E2E ASR system.

We introduce VoiceFilter-Lite, a single-channel source separation model that runs on the device to preserve only the speech signals from a target user, as part of a streaming speech recognition system.Delivering such a model presents numerous challenges: It should improve the performance when the input signal consists of overlapped speech, and must not hurt the speech recognition performance under all other acoustic conditions.Besides, this model must be tiny, fast, and perform inference in a streaming fashion, in order to have minimal impact on CPU, memory, battery and latency.We propose novel techniques to meet these multi-faceted requirements, including using a new asymmetric loss, and adopting adaptive runtime suppression strength.We also show that such a model can be quantized as a 8-bit integer model and run in realtime.

The tradeoff between word error rate (WER) and latency is very important for streaming automatic speech recognition (ASR) applications. We want the system to endpoint and close the microphone as quickly as possible, without degrading WER. Conventional ASR systems rely on a separately trained endpointing module, which interacts with the acoustic, pronunciation and language model (AM, PM, and LM) components, and can result in a higher WER or a larger latency. In going with the all-neural spirit of end-to-end (E2E) models, which fold the AM, PM and LM into a single neural network, in this work we look at folding the endpointer into this E2E model to assist with the endpointing task. We refer to this jointly optimized model - which performs both recognition and endpointing - as an E2E enpointer. On a large vocabulary Voice Search task, we show that the combination of such an E2E endpoiner with a conventional endpointer results in no quality degradation, while reducing latency by more than a factor of 2 compared to using a separate endpointer with the E2E model.

End-to-end (E2E) models, which directly predict output character sequences given input speech, are good candidates for on-device speech recognition. E2E models, however, present numerous challenges: In order to be truly useful, such models must decode speech utterances in a streaming fashion, in real time; they must be robust to the long tail of use cases; they must be able to leverage user-specific context (e.g., contact lists); and above all, they must be extremely accurate. In this work, we describe our efforts at building an E2E speech recog-nizer using a recurrent neural network transducer. In experimental evaluations, we find that the proposed approach can outperform a conventional CTC-based model in terms of both latency and accuracy in a number of evaluation categories.

The requirements for many applications of state-of-the-art speech recognition systems include not only low word error rate (WER) but also low latency. Specifically, for many use-cases, the system must be able to decode utterances in a streaming fashion and faster than real-time. Recently, a streaming recurrent neural network transducer (RNN-T) end-to-end (E2E) model has shown to be a good candidate for on-device speech recognition, with improved WER and latency metrics compared to conventional on-device models [1]. However, this model still lags behind a large state-of-the-art conventional model in quality [2]. On the other hand, a non-streaming E2E Listen, Attend and Spell (LAS) model has shown comparable quality to large conventional models [3]. This work aims to bring the quality of an E2E streaming model closer to that of a conventional system by incorporating a LAS network as a second-pass component, while still abiding by latency constraints. Our proposed two-pass model achieves a 17%-22% relative reduction in WER compared to RNN-T alone and increases latency by a small fraction over RNN-T.

We develop streaming keyword spotting systems using a recurrent neural network transducer (RNN-T) model: an all-neural, end-to-end trained, sequence-to-sequence model which jointly learns acoustic and language model components. Our models are trained to predict either phonemes or graphemes as subword units, thus allowing us to detect arbitrary keyword phrases, without any out-of-vocabulary words. In order to adapt the models to the requirements of keyword spotting, we propose a novel technique which biases the RNN-T system towards a specific keyword of interest. Our systems are compared against a strong sequence-trained, connectionist temporal classification (CTC) based “keyword-filler” baseline, which is augmented with a separate phoneme language model. Overall, our RNN-T system with the proposed biasing technique significantly improves performance over the baseline system.

Out-of-vocabulary (OOV) keywords present a challenge for keyword search (KWS) systems especially in the low-resource setting. Previous research has centered around approaches that use a variety of subword units to recover OOV words. This paper systematically investigates morphology-based subword modeling approaches on seven low-resource languages. We show that using morphological subword units (morphs) in speech recognition decoding is substantially better than expanding word-decoded lattices into subword units including phones, syllables and morphs. As alternatives to grapheme-based morphs, we apply unsupervised morphology learning to sequences of phonemes, graphones, and syllables. Using one of these phone-based morphs is almost always better than using the grapheme-based morphs, but the particular choice varies with the language. By combining the different methods, a substantial gain is obtained over the best single case for all languages, especially for OOV performance.

Automatic Speech Recognition systems suffer from severe performance degradation in the presence of myriad complicating factors such as noise, reverberation, multiple speech sources, multiple recording devices, etc. Previous challenges have sparked much innovation when it comes to designing systems capable of handling these complications. In this spirit, the CHiME-3 challenge presents system builders with the task of recognizing speech in a real-world noisy setting wherein speakers talk to an array of 6 microphones in a tablet. In order to address these issues, we explore the effectiveness of first applying a model-based source separation mask to the output of a beamformer that combines the source signals recorded by each microphone, followed by a DNN-based front end spectral mapper that predicts clean filterbank features. The source separation algorithm MESSL (Model-based EM Source Separation and Localization) has been extended from two channels to multiple channels in order to meet the demands of the challenge. We report on interactions between the two systems, cross-cut by the use of a robust beamforming algorithm called BeamformIt. Evaluations of different system settings reveal that combining MESSL and the spectral mapper together on the baseline beamformer algorithm boosts the performance substantially.

This paper investigates a weighted finite state transducer (WFST) based syllable decoding and transduction method for keyword search (KWS), and compares it with sub-word search and phone confusion methods in detail. Acoustic context dependent phone models are trained from word forced alignments and then used for syllable decoding and lattice generation. Out-of-vocabulary (OOV) keyword pronunciations are produced using a grapheme-to-syllable (G2S) system and then used to construct a lexical transducer. The lexical transducer is then composed with a keyword-boosted language model (LM) to transduce the syllable lattices to word lattices for final KWS. Word Error Rates (WER) and KWS results are reported for 5 different languages. It is shown that the syllable transduction method gives comparable KWS results to the syllable search and phone confusion methods. Combination of these three methods further improves OOV KWS performance.

Automatic speech recognition (ASR) systems suffer from performance degradation under noisy and reverberant conditions. In this work, we explore a deep neural network (DNN) based approach for spectral feature mapping from corrupted speech to clean speech. The DNN based mapping substantially reduces interference and produces estimated clean spectral features for ASR training and decoding. We experiment with several different feature mapping approaches and demonstrate that a DNN trained to predict clean log filterbank coefficients from noisy spectrogram directly can be extremely effective. The experiments show that the ASR systems with these cleaned features perform well under joint noisy and reverberant conditions, and achieve the state-of-the-art results on the CHiME-2 corpus with stereo (corrupted and clean) data.

Discriminative segmental models, such as segmental conditional random fields (SCRFs), have been successfully applied to speech recognition recently in lattice rescoring to integrate detectors across different levels of units, such as phones and words. However, the lattice generation has been constrained by a baseline decoder, typically a frame-based hybrid HMMDNN system, which still suffers from the well-known frame independent assumption. In this paper, we propose to use SCRFs with DNNs directly as the acoustic model, a one-pass unified framework that can utilize local phone classifiers, phone transitions and long-span features, in direct word decoding to model phones or sub-phonetic segments with variable length. We describe a WFST-based approach to utilize the proposed acoustic model efficiently with the language model in first-pass word recognition. Our evaluation on the WSJ corpus shows our SCRF-DNN system outperforms a hybrid HMM-DNN system and a frame-level CRF-DNN system using the same label space.

With the rapid development of big data and cloud computing, big data analytics as a service in the cloud is becoming increasingly popular. More and more individuals and organizations tend to rent virtual cluster to store and analyze data rather than building their own data centers. However, in virtualization environment, whether scaling out using a cluster with more nodes to process big data is better than scaling up by adding more resources to the original virtual machines (VMs) in cluster is not clear. In this paper, we study the scalability performance issues of hadoop virtual cluster with cost consideration. We first present the design and implementation of VirtualMR platform which can provide users with scalable hadoop virtual cluster services for the MapReduce based big data analytics. Then we run a series of hadoop benchmarks and real parallel machine learning algorithms to evaluate the scalability performance, including scale-up method and scale-out method. Finally, we integrate our platform with resource monitoring module and propose a system tuner. By analyzing the monitored data, we dynamically adjust the parameters of hadoop framework and virtual machine configuration to improve resource utilization and reduce rent cost. Experimental results show that the scale-up method outperforms the scale-out method for CPU-bound applications, and it is opposite for I/O-bound applications. The results also verify the efficiency of system tuner to increase resource utilization and reduce rent cost.

This work compares ASR decoding at different subword levels crossed with alternative keyword search strategies to handle the OOV issue for keyword spotting in the low-resource setting. We show that a morpheme-based subword modeling approach is effective in recovering OOV keywords within a Turkish low-resource keyword spotting task, where mixed word and morpheme decoding approach outperforms the traditional subword-based search from word-decoded lattices that are broken down to subword lattices. Furthermore, unsupervised learning of morphology works almost as well as a rule-based system designed for the language despite the low-resource condition. A staged keyword search strategy benefits from both methods of morphological analysis.

This paper presents a weighted finite state transducer (WFST) based syllable decoding and transduction framework for keyword search (KWS). Acoustic context dependent phone models are trained from word forced alignments. Then syllable decoding is done with lattices generated using a syllable lexicon and language model (LM). To process out-of-vocabulary (OOV) keywords, pronunciations are produced using a grapheme-to-syllable (G2S) system. A syllable to word lexical transducer containing both in-vocabulary (IV) and OOV keywords is then constructed and composed with a keyword-boosted LM transducer. The composed transducer is then used to transduce syllable lattices to word lattices for final KWS. We show that our method can effectively perform KWS on both IV and OOV keywords, and yields up to 0.03 Actual Term-Weighted Value (ATWV) improvement over searching keywords directly in subword lattices. Word Error Rates (WER) and KWS results are reported for three different languages.

Conditional random fields (CRFs) are probabilistic sequence models that have been applied in the last decade to a number of applications in audio, speech, and language processing. In this paper, we provide a tutorial overview of CRF technologies, pointing to other resources for more in-depth discussion; in particular, we describe the common linear-chain model as well as a number of common extensions within the CRF family of models. An overview of the mathematical techniques used in training and evaluating these models is also provided, as well as a discussion of the relationships with other probabilistic models. Finally, we survey recent work in speech, audio, and language processing to show how the same CRF technology can be deployed in different scenarios.