In the era of large models, the autoregressive nature of decoding often results in latency serving as a significant bottleneck. We propose a non-autoregressive LM-fused ASR system that effectively leverages the parallelization capabilities of accelerator hardware. Our approach combines the Universal Speech Model (USM) and the PaLM 2 language model in per-segment scoring mode, achieving an average relative WER improvement across all languages of 10.8% on FLEURS and 3.6% on YouTube captioning. Furthermore, our comprehensive ablation study analyzes key parameters such as LLM size, context length, vocabulary size, fusion methodology. For instance, we explore the impact of LLM size ranging from 128M to 340B parameters on ASR performance. This study provides valuable insights into the factors influencing the effectiveness of practical large-scale LM-fused speech recognition systems.

ASR models often suffer from a long-form deletion problem where the model predicts sequential blanks instead of words when transcribing a lengthy audio (in the order of minutes or hours). From the perspective of a user or downstream system consuming the ASR results, this behavior can be perceived as the model “being stuck”, and potentially make the product hard to use. One of the culprits for long-form deletion is training-test data mismatch, which can happen even when the model is trained on diverse and large-scale data collected from multiple application domains. In this work, we introduce a novel technique to simultaneously model different groups of speakers in the audio along with the standard transcript tokens. Speakers are grouped as primary and non-primary, which connects the application domains and significantly alleviates the long-form deletion problem. This improved model neither needs any additional training data nor incurs additional training or inference cost.

This paper introduces a novel voice conversion (VC) model, guided by text instructions such as “articulate slowly with a deep tone“ or “speak in a cheerful boyish voice”. Unlike traditional methods that rely on reference utterances to determine the attributes of the converted speech, using text instruction adds versatility and specificity to voice conversion. The proposed VC model is a neural codec language model which processes a sequence of discrete codes, resulting in the code sequence of converted speech. It utilizes text instructions as style prompts to modify the prosody and emotional information of the given speech. In contrast to previous approaches, which often rely on employing separate encoders like prosody and content encoders to handle different aspects of the source speech, our model handles various information of speech in an end-to-end manner. Experiments have demonstrated the impressive capabilities of our model in comprehending instructions and delivering reasonable results1.

Typical automatic speech recognition (ASR) systems are built to recognize independent utterances without using the cross-utterance context. However, the context over multiple utterances often provides useful information for the ASR task. In this work, we propose a context-aware end-to-end ASR model that injects the self-attentive context embedding into the decoder of the recurrent neural network transducer (RNN-T). We also propose a factorised 3-way tensor fusion approach to fuse the context embedding with the acoustic representations extracted from the acoustic encoder and the text representations obtained using the prediction network based on the previous subword units. Experimental results on a long-form Youtube ASR task shows that the proposed approach achieves 10.8% relative word error rate reductions.

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.

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.

Word-piece models (WPMs) are commonly used subword units in state-of-the-art end-to-end automatic speech recognition (ASR) systems. For multilingual ASR, due to the differences in written scripts across languages, multilingual WPMs bring the challenges of having overly large output layers and scaling to more languages. In this work, we propose a universal monolingual output layer (UML) to address such problems. Instead of one output node for only one WPM, UML re-associates each output node with multiple WPMs, one for each language, and results in a smaller monolingual output layer shared across languages. Consequently, the UML enables to switch in the interpretation of each output node depending on the language of the input speech. Experimental results on an 11-language voice search task demonstrated the feasibility of using UML for high-quality and high-efficiency multilingual streaming ASR.

In this work, we introduce a "score-based assessment" framework for estimating the transferability of pre-trained speech models (PSMs) for fine-tuning target tasks.We leverage upon two representation theories, Bayesian likelihood estimation and optimal transport, to generate rank scores for the PSM candidates using the extracted representations.Our framework efficiently computes transferability scores without actual finetuning of candidate models or layers by making a temporal independent hypothesis.We evaluate some popular supervised speech models (e.g., Conformer RNN-Transducer) and selfsupervised speech models (e.g., HuBERT) in cross-layer and cross-model settings using public data.Experimental results show a high Spearman's rank correlation and low p-value between our estimation framework and fine-tuning ground truth.Our proposed transferability framework requires less computational time and resources, making it a resource-saving and timeefficient approach for tuning speech foundation models.

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.

The recurrent neural network transducer (RNN-T) has recently become the mainstream end-to-end approach for streaming automatic speech recognition (ASR). To estimate the output distributions over subword units, RNN-T uses a fully connected layer as the joint network to fuse the acoustic representations extracted using the acoustic encoder with the text representations obtained using the prediction network based on the previous subword units. In this paper, we propose to use gating, bilinear pooling, and a combination of them in the joint network to produce more expressive representations to feed into the output layer. A regularisation method is also proposed to enable better acoustic encoder training by reducing the gradients back-propagated into the prediction network at the beginning of RNN-T training. Experimental results on a multilingual ASR setting for voice search over nine languages show that the joint use of the proposed methods can result in 4%–5% relative word error rate reductions with only a few million extra parameters.

In voice-enabled applications, a predetermined hotword is usually used to activate a device in order to attend to the query.However, speaking queries followed by a hotword each time introduces a cognitive burden in continued conversations.To avoid repeating a hotword, we propose a streaming end-to-end (E2E) intended query detector that identifies the utterances directed towards the device and filters out other utterances not directed towards device.The proposed approach incorporates the intended query detector into the E2E model that already folds different components of the speech recognition pipeline into one neural network.The E2E modeling on speech decoding and intended query detection also allows us to declare a quick intended query detection based on early partial recognition result, which is important to decrease latency and make the system responsive.We demonstrate that the proposed E2E approach yields a 22% relative improvement on equal error rate (EER) for the detection accuracy and 600 ms latency improvement compared with an independent intended query detector.In our experiment, the proposed model detects whether the user is talking to the device with a 8.7% EER within 1.4 seconds of median latency after user starts speaking.

Improving the performance of end-to-end ASR models on long utterances ranging from minutes to hours in length is an ongoing challenge in speech recognition. A common solution is to segment the audio in advance using a separate voice activity detector (VAD) that decides segment boundary locations based purely on acoustic speech/non-speech information. VAD segmenters, however, may be sub-optimal for real-world speech where, e.g., a complete sentence that should be taken as a whole may contain hesitations in the middle ("set an alarm for... 5 o'clock"). We propose to replace the VAD with an end-to-end ASR model capable of predicting segment boundaries in a streaming fashion, allowing the segmentation decision to be conditioned not only on better acoustic features but also on semantic features from the decoded text with negligible extra computation. In experiments on real world long-form audio (YouTube) with lengths of up to 30 minutes, we demonstrate 8.5% relative WER improvement and 250 ms reduction in median end-of-segment latency compared to the VAD segmenter baseline on a state-of-the-art Conformer RNN-T model.

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.

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).

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.

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.

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.

In speech recognition systems, we generally differentiate the role of endpointers between long-form speech and voice queries, where they are responsible for speech detection and query endpoint detection respectively. Detection of speech is useful for segmentation and pre-filtering in long-form speech processing. On the other hand, query endpoint detection predicts when to stop listening and send audio received so far for actions. It thus determines system latency and is an essential component for interactive voice systems. For both tasks, endpointer needs to be robust in challenging environments, including noisy conditions, reverberant environments and environments with background speech, and it has to generalize well to different domains with different speaking styles and rhythms. This work investigates building a unified endpointer by folding the separate speech detection and query endpoint detection tasks into a single neural network model through multitask learning. A categorical domain representation is further incorporated into the model to encourage learning domain specific information. The final unified model achieves around 100 ms (18% relatively) latency improvement for near-field voice queries and 150 ms (21% relatively) for far-field voice queries over simply pooling all the data together and 7% relative frame error rate reduction for long-form speech compared to a standalone speech detection model. The proposed approach also shows good robustness to noisy environments and yields 180 ms latency improvement on voice queries from an unseen domain.

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.

Voice activity detection (VAD) is the task of predicting which parts of an utterance contains speech versus background noise. It is an important first step to determine which samples to send to the decoder and when to close the microphone. The long short-term memory neural network (LSTM) is a popular architecture for sequential modeling of acoustic signals, and has been successfully used in several VAD applications. However, it has been observed that LSTMs suffer from state saturation problems when the utterance is long (i.e., for voice dictation tasks), and thus requires the LSTM state to be periodically reset. In this paper, we propose an alternative architecture that does not suffer from saturation problems by modeling temporal variations through a stateless dilated convolution neural network (CNN). The proposed architecture differs from conventional CNNs in three respects: it uses dilated causal convolution, gated activations and residual connections. Results on a Google Voice Typing task shows that the proposed architecture achieves 14% relative FA improvement at a FR of 1% over state-of-the-art LSTMs for VAD task. We also include detailed experiments investigating the factors that distinguish the proposed architecture from conventional convolution.

A large body of research has shown that acoustic features for speech recognition can be learned from data using neural networks with multiple hidden layers (DNNs) and that these learned features are superior to standard features (e.g., MFCCs). However, this superiority is usually demonstrated when the data used to learn the features is very similar in character to the data used to test recognition performance. An open question is how well these learned features generalize to realistic data that is different in character to their training data. The ability of a feature representation to generalize to unfamiliar data is a highly desirable form of robustness. In this paper we investigate the robustness of two DNN-based feature sets to training/test mismatch using the ICSI meeting corpus. The experiments were performed under 3 training/test scenarios: (1) matched near-field (2) matched far-field and (3) the mismatched condition near-field training with far-field testing. The experiments leverage simulation and a novel sampling process that we have developed for diagnostic analysis within the HMM-based speech recognition framework. First, diagnostic analysis shows that a DNN-based feature representation that uses MFCC inputs (MFCC-DNN) is indeed superior to the corresponding MFCC baselines in the two matched scenarios where the source of recognition errors are from incorrect model, but the DNN-based features and MFCCs have nearly identical and poor performance in the mismatched scenario. Second, we show that a DNN-based feature representation that uses a more robust input, namely power normalized spectrum (PNS) and Gabor filters, performs nearly as well as the MFCC-DNN features in the matched scenarios and much better than MFCCs and MFCC-DNNs in the mismatched scenario.

As has been extensively shown, acoustic features for speech recognition can be learned from neural networks with multiple hidden layers. However, the learned transformations may not sufficiently generalize to test sets that have a significant mismatch to the training data. Gabor features, on the other hand, are generated from spectro-temporal filters designed to model human auditory processing. In previous work, these features are used as inputs to neural networks, which improved word accuracy for speech recognition in the presence of noise. Here we propose a neural network architecture called a Gabor Convolutional Neural Network (GCNN) that incorporates Gabor functions into convolutional filter kernels. In this architecture, a variety of Gabor features served as the multiple feature maps of the convolutional layer. The filter coefficients are further tuned by back-propagation training. Experiments used two noisy versions of the WSJ corpus: Aurora 4, and RATS re-noised WSJ. In both cases, the proposed architecture performs better than other noise-robust features that we have tried, namely, ETSI-AFE, PNCC, Gabor features without the CNN-based approach, and our best neural network features that don’t incorporate Gabor functions.

Previous work has demonstrated that spectro-temporal Gabor features reduced word error rates for automatic speech recognition under noisy conditions. However, the features based on mel spectra were easily corrupted in the presence of noise or channel distortion. We have exploited an algorithm for power normalized cepstral coefficients (PNCCs) to generate a more robust spectro-temporal representation. We refer to it as power normalized spectrum (PNS), and to the corresponding output processed by Gabor filters and MLP nonlinear weighting as PNS-Gabor. We show that the proposed feature outperforms state-of-the-art noise-robust features, ETSI-AFE and PNCC for both Aurora2 and a noisy version of the Wall Street Jounal (WSJ) corpus. A comparison of the individual processing steps of mel spectra and PNS shows that power bias subtraction is the most important aspect of PNS-Gabor features to provide an improvement over Mel-Gabor features. The result indicates that Gabor processing compensates the limitation of PNCC for channels with frequency-shift characteristic. Overall, PNS-Gabor features decrease the word error rate by 32% relative to MFCC and 13% relative to PNCC in Aurora2. For noisy WSJ, they decrease the word error rate by 30.9% relative to MFCC and 24.7% relative to PNCC.

Given a test waveform, state-of-the-art ASR systems extract a sequence of MFCC features and decode them with a set of trained HMMs. When this test data is clean, and it matches the condition used for training the models, then there are few errors. While it is known that ASR systems are brittle in noisy or mismatched conditions, there has been little work in quantitatively attributing the errors to features or to models. This paper attributes the sources of these errors in three conditions: (a) matched near-field, (b) matched far-field, and a (c) mismatched condition. We undertake a series of diagnostic analyses employing the bootstrap method to probe a meeting room ASR system. Results show that when the conditions are matched (even if they are far-field), the model errors dominate; however, in mismatched conditions features are neither invariant nor separable and this causes as many errors as the model does.

The outputs of multi-layer perceptron (MLP) classifiers have been successfully used in tandem systems as features for HMM-based automatic speech recognition. In a previous paper, we proposed data-driven clustered hierarchical MLP (CHMLP) tandem system yielding improved performance by dividing the complicated global phone classification problem into simpler hierarchical tasks, in which specialized MLPs are trained to classify small clusters of confusing phones in a hierarchical structure. In this paper a bottom-up processing is further proposed to enhance the classification in the above CHMLP and offer even better performance. MLP rescoring for the tandem system is also investigated. The best result achieved 19.1% relative error reduction over the MFCC baseline.

This paper investigates the use of a hierarchy of Neural Networks for performing data driven feature extraction.Two different hierarchical structures based on long and short temporal context are considered.Features are tested on two different LVCSR systems for Meetings data (RT05 evaluation data) and for Arabic Broadcast News (BNAT05 evaluation data).The hierarchical NNs outperforms the single NN features consistently on different type of data and tasks and provides significant improvements w.r.t.respective baselines systems.Best results are obtained when different time resolutions are used at different level of the hierarchy.