Duc Le

Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri et al. · allen-ai, berkeley

Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.

Paden Tomasello, Akshat Shrivastava, Daniel Lazar et al. · meta-ai, uw

End-to-end spoken language understanding (SLU) predicts intent directly from audio using a single model. It promises to improve the performance of assistant systems by leveraging acoustic information lost in the intermediate textual representation and preventing cascading errors from Automatic Speech Recognition (ASR). Further, having one unified model has efficiency advantages when deploying assistant systems on-device. However, the limited number of public audio datasets with semantic parse labels hinders the research progress in this area. In this paper, we release the Spoken Task-Oriented semantic Parsing (STOP) dataset, the largest and most complex SLU dataset to be publicly available. Additionally, we define low-resource splits to establish a benchmark for improving SLU when limited labeled data is available. Furthermore, in addition to the human-recorded audio, we are releasing a TTS-generated version to benchmark the performance for low-resource domain adaptation of end-to-end SLU systems. Initial experimentation show end-to-end SLU models performing slightly worse than their cascaded counterparts, which we hope encourages future work in this direction.

Duc Le, Akshat Shrivastava, Paden Tomasello et al. · meta-ai

We propose a novel deliberation-based approach to end-to-end (E2E) spoken language understanding (SLU), where a streaming automatic speech recognition (ASR) model produces the first-pass hypothesis and a second-pass natural language understanding (NLU) component generates the semantic parse by conditioning on both ASR's text and audio embeddings. By formulating E2E SLU as a generalized decoder, our system is able to support complex compositional semantic structures. Furthermore, the sharing of parameters between ASR and NLU makes the system especially suitable for resource-constrained (on-device) environments; our proposed approach consistently outperforms strong pipeline NLU baselines by 0.60% to 0.65% on the spoken version of the TOPv2 dataset (STOP). We demonstrate that the fusion of text and audio features, coupled with the system's ability to rewrite the first-pass hypothesis, makes our approach more robust to ASR errors. Finally, we show that our approach can significantly reduce the degradation when moving from natural speech to synthetic speech training, but more work is required to make text-to-speech (TTS) a viable solution for scaling up E2E SLU.

Suyoun Kim, Akshat Shrivastava, Duc Le et al. · meta-ai

End-to-end (E2E) spoken language understanding (SLU) systems that generate a semantic parse from speech have become more promising recently. This approach uses a single model that utilizes audio and text representations from pre-trained speech recognition models (ASR), and outperforms traditional pipeline SLU systems in on-device streaming scenarios. However, E2E SLU systems still show weakness when text representation quality is low due to ASR transcription errors. To overcome this issue, we propose a novel E2E SLU system that enhances robustness to ASR errors by fusing audio and text representations based on the estimated modality confidence of ASR hypotheses. We introduce two novel techniques: 1) an effective method to encode the quality of ASR hypotheses and 2) an effective approach to integrate them into E2E SLU models. We show accuracy improvements on STOP dataset and share the analysis to demonstrate the effectiveness of our approach.

Mu Yang, Andros Tjandra, Chunxi Liu et al.

Neural network pruning compresses automatic speech recognition (ASR) models effectively. However, in multilingual ASR, language-agnostic pruning may lead to severe performance drops on some languages because language-agnostic pruning masks may not fit all languages and discard important language-specific parameters. In this work, we present ASR pathways, a sparse multilingual ASR model that activates language-specific sub-networks ("pathways"), such that the parameters for each language are learned explicitly. With the overlapping sub-networks, the shared parameters can also enable knowledge transfer for lower-resource languages via joint multilingual training. We propose a novel algorithm to learn ASR pathways, and evaluate the proposed method on 4 languages with a streaming RNN-T model. Our proposed ASR pathways outperform both dense models and a language-agnostically pruned model, and provide better performance on low-resource languages compared to the monolingual sparse models.

Andros Tjandra, Nayan Singhal, David Zhang et al.

End-to-end multilingual ASR has become more appealing because of several reasons such as simplifying the training and deployment process and positive performance transfer from high-resource to low-resource languages. However, scaling up the number of languages, total hours, and number of unique tokens is not a trivial task. This paper explores large-scale multilingual ASR models on 70 languages. We inspect two architectures: (1) Shared embedding and output and (2) Multiple embedding and output model. In the shared model experiments, we show the importance of tokenization strategy across different languages. Later, we use our optimal tokenization strategy to train multiple embedding and output model to further improve our result. Our multilingual ASR achieves 13.9%-15.6% average WER relative improvement compared to monolingual models. We show that our multilingual ASR generalizes well on an unseen dataset and domain, achieving 9.5% and 7.5% WER on Multilingual Librispeech (MLS) with zero-shot and finetuning, respectively.

Thinh Phan, Duc Le, Patel Brijesh et al.

Electrocardiogram (ECG) signal is one of the most effective sources of information mainly employed for the diagnosis and prediction of cardiovascular diseases (CVDs) connected with the abnormalities in heart rhythm. Clearly, single modality ECG (i.e. time series) cannot convey its complete characteristics, thus, exploiting both time and time-frequency modalities in the form of time-series data and spectrogram is needed. Leveraging the cutting-edge self-supervised learning (SSL) technique on unlabeled data, we propose SSL-based multimodality ECG classification. Our proposed network follows SSL learning paradigm and consists of two modules corresponding to pre-stream task, and down-stream task, respectively. In the SSL-pre-stream task, we utilize self-knowledge distillation (KD) techniques with no labeled data, on various transformations and in both time and frequency domains. In the down-stream task, which is trained on labeled data, we propose a gate fusion mechanism to fuse information from multimodality.To evaluate the effectiveness of our approach, ten-fold cross validation on the 12-lead PhysioNet 2020 dataset has been conducted.

Sung-Lin Yeh, Wei Zhou, Gil Keren et al.

Recent speech language models rely on encoders that are optimized separately from autoregressive models. Since these encoders are unaware of the downstream objectives, the extracted representations may not be optimal for downstream tasks. To address this limitation, we introduce a discrete latent variable model on mel spectrograms that jointly optimizes the encoder and the speech language model. Joint optimization not only brings improvements over codec-based and other mel-spectrogram-based baselines on zero-shot Text-to-Speech (TTS) and Speech-to-Text (STT) tasks, but also effectively alleviates common issues in autoregressive mel-spectrogram modeling, such as prolonged silence generation and word omissions.

Suyoun Kim, Ke Li, Lucas Kabela et al.

Recently, there has been an increasing interest in two-pass streaming end-to-end speech recognition (ASR) that incorporates a 2nd-pass rescoring model on top of the conventional 1st-pass streaming ASR model to improve recognition accuracy while keeping latency low. One of the latest 2nd-pass rescoring model, Transformer Rescorer, takes the n-best initial outputs and audio embeddings from the 1st-pass model, and then choose the best output by re-scoring the n-best initial outputs. However, training this Transformer Rescorer requires expensive paired audio-text training data because the model uses audio embeddings as input. In this work, we present our Joint Audio/Text training method for Transformer Rescorer, to leverage unpaired text-only data which is relatively cheaper than paired audio-text data. We evaluate Transformer Rescorer with our Joint Audio/Text training on Librispeech dataset as well as our large-scale in-house dataset and show that our training method can improve word error rate (WER) significantly compared to standard Transformer Rescorer without requiring any extra model parameters or latency.

Duc Le, Panos P. Markopoulos

Singular-Value Decomposition (SVD) is a ubiquitous data analysis method in engineering, science, and statistics. Singular-value estimation, in particular, is of critical importance in an array of engineering applications, such as channel estimation in communication systems, electromyography signal analysis, and image compression, to name just a few. Conventional SVD of a data matrix coincides with standard Principal-Component Analysis (PCA). The L2-norm (sum of squared values) formulation of PCA promotes peripheral data points and, thus, makes PCA sensitive against outliers. Naturally, SVD inherits this outlier sensitivity. In this work, we present a novel robust non-parametric method for SVD and singular-value estimation based on a L1-norm (sum of absolute values) formulation, which we name L1-cSVD. Accordingly, the proposed method demonstrates sturdy resistance against outliers and can facilitate more reliable data analysis and processing in a wide range of engineering applications.

Julian D. Parker, Janne Spijkervet, Katerina Kosta et al.

End-to-end generation of musical audio using deep learning techniques has seen an explosion of activity recently. However, most models concentrate on generating fully mixed music in response to abstract conditioning information. In this work, we present an alternative paradigm for producing music generation models that can listen and respond to musical context. We describe how such a model can be constructed using a non-autoregressive, transformer-based model architecture and present a number of novel architectural and sampling improvements. We train the described architecture on both an open-source and a proprietary dataset. We evaluate the produced models using standard quality metrics and a new approach based on music information retrieval descriptors. The resulting model reaches the audio quality of state-of-the-art text-conditioned models, as well as exhibiting strong musical coherence with its context.

Cuong Le, Huy-Phuong Le, Duc Le et al.

Body dynamics are crucial information for the analysis of human motions in important research fields, ranging from biomechanics, sports science to computer vision and graphics. Modern approaches collect the body dynamics, external reactive force specifically, via force plates, synchronizing with human motion capture data, and learn to estimate the dynamics from a black-box deep learning model. Being specialized devices, force plates can only be installed in laboratory setups, imposing a significant limitation on the learning of human dynamics. To this end, we propose a novel method for estimating human ground reaction dynamics directly from the more reliable motion capture data with physics laws and computational simulation as constrains. We introduce a highly accurate and robust method for computing ground reaction forces from motion capture data using Euler's integration scheme and PD algorithm. The physics-based reactive forces are used to inform the learning model about the physics-informed motion dynamics thus improving the estimation accuracy. The proposed approach was tested on the GroundLink dataset, outperforming the baseline model on: 1) the ground reaction force estimation accuracy compared to the force plates measurement; and 2) our simulated root trajectory precision. The implementation code is available at https://github.com/cuongle1206/Phys-GRD

Matthew Perez, Duc Le, Amrit Romana et al.

Paraphasias are speech errors that are often characteristic of aphasia and they represent an important signal in assessing disease severity and subtype. Traditionally, clinicians manually identify paraphasias by transcribing and analyzing speech-language samples, which can be a time-consuming and burdensome process. Identifying paraphasias automatically can greatly help clinicians with the transcription process and ultimately facilitate more efficient and consistent aphasia assessment. Previous research has demonstrated the feasibility of automatic paraphasia detection by training an automatic speech recognition (ASR) model to extract transcripts and then training a separate paraphasia detection model on a set of hand-engineered features. In this paper, we propose a novel, sequence-to-sequence (seq2seq) model that is trained end-to-end (E2E) to perform both ASR and paraphasia detection tasks. We show that the proposed model outperforms the previous state-of-the-art approach for both word-level and utterance-level paraphasia detection tasks and provide additional follow-up evaluations to further understand the proposed model behavior.

Yen-Ju Lu, Yashesh Gaur, Wei Zhou et al.

Auto-regressive speech-text models are typically pre-trained on a large number of interleaved sequences of text tokens and raw speech encoded as speech tokens using vector quantization. These models have demonstrated state-of-the-art performance in speech-to-speech understanding and generation benchmarks, together with promising scaling laws, primarily enabled by the representational alignment between text and speech. Nevertheless, they suffer from shortcomings, partly owing to the disproportionately longer sequences of speech tokens in contrast to textual tokens. This results in a large compute imbalance between modalities during pre-training as well as during inference, and a potential hindrance to effectively aligning speech and text, ultimately translating to several orders of magnitude slower scaling laws. We introduce the Latent Speech-Text Transformer (LST), which makes pre-training speech-text models more data-efficient by dynamically and inexpensively aggregating speech tokens into latent speech patches. These patches serve as higher-level units that can either align with corresponding textual units to aid capability transfer or even encapsulate common speech sequences like silences to be more compute-efficient. We show that LST outperforms vanilla approaches on speech-to-speech as well as text-to-text benchmarks in both data- and compute-controlled settings, the former indicating more effective representational alignment and the latter indicating steeper scaling laws for speech-text models. On HellaSwag story completion, LST achieves 6.5% absolute gain in speech accuracy under compute-controlled training and 5.3% under data-controlled training, while also improving text performance. We will release our models, code, and the evaluation data to facilitate further research.

Tung Luu, Nam Le, Duc Le et al.

Visual counterfactual explanations are ideal hypothetical images that change the decision-making of the classifier with high confidence toward the desired class while remaining visually plausible and close to the initial image. In this paper, we propose a new approach to tackle two key challenges in recent prominent works: i) determining which specific counterfactual features are crucial for distinguishing the "concept" of the target class from the original class, and ii) supplying valuable explanations for the non-robust classifier without relying on the support of an adversarially robust model. Our method identifies the essential region for modification through algorithms that provide visual explanations, and then our framework generates realistic counterfactual explanations by combining adversarial attacks based on pruning the adversarial gradient of the target classifier and the latent diffusion model. The proposed method outperforms previous state-of-the-art results on various evaluation criteria on ImageNet and CelebA-HQ datasets. In general, our method can be applied to arbitrary classifiers, highlight the strong association between visual and counterfactual explanations, make semantically meaningful changes from the target classifier, and provide observers with subtle counterfactual images.

Trang Le, Daniel Lazar, Suyoun Kim et al.

Spoken Language Understanding (SLU) is a critical component of voice assistants; it consists of converting speech to semantic parses for task execution. Previous works have explored end-to-end models to improve the quality and robustness of SLU models with Deliberation, however these models have remained autoregressive, resulting in higher latencies. In this work we introduce PRoDeliberation, a novel method leveraging a Connectionist Temporal Classification-based decoding strategy as well as a denoising objective to train robust non-autoregressive deliberation models. We show that PRoDeliberation achieves the latency reduction of parallel decoding (2-10x improvement over autoregressive models) while retaining the ability to correct Automatic Speech Recognition (ASR) mistranscriptions of autoregressive deliberation systems. We further show that the design of the denoising training allows PRoDeliberation to overcome the limitations of small ASR devices, and we provide analysis on the necessity of each component of the system.

Zhuangqun Huang, Gil Keren, Ziran Jiang et al.

Aiming at reducing the reliance on expensive human annotations, data synthesis for Automatic Speech Recognition (ASR) has remained an active area of research. While prior work mainly focuses on synthetic speech generation for ASR data augmentation, its combination with text generation methods is considerably less explored. In this work, we explore text augmentation for ASR using large-scale pre-trained neural networks, and systematically compare those to traditional text augmentation methods. The generated synthetic texts are then converted to synthetic speech using a text-to-speech (TTS) system and added to the ASR training data. In experiments conducted on three datasets, we find that neural models achieve 9%-15% relative WER improvement and outperform traditional methods. We conclude that text augmentation, particularly through modern neural approaches, is a viable tool for improving the accuracy of ASR systems.

Jay Mahadeokar, Yangyang Shi, Ke Li et al.

Streaming ASR with strict latency constraints is required in many speech recognition applications. In order to achieve the required latency, streaming ASR models sacrifice accuracy compared to non-streaming ASR models due to lack of future input context. Previous research has shown that streaming and non-streaming ASR for RNN Transducers can be unified by cascading causal and non-causal encoders. This work improves upon this cascaded encoders framework by leveraging two streaming non-causal encoders with variable input context sizes that can produce outputs at different audio intervals (e.g. fast and slow). We propose a novel parallel time-synchronous beam search algorithm for transducers that decodes from fast-slow encoders, where the slow encoder corrects the mistakes generated from the fast encoder. The proposed algorithm, achieves up to 20% WER reduction with a slight increase in token emission delays on the public Librispeech dataset and in-house datasets. We also explore techniques to reduce the computation by distributing processing between the fast and slow encoders. Lastly, we explore sharing the parameters in the fast encoder to reduce the memory footprint. This enables low latency processing on edge devices with low computation cost and a low memory footprint.

Antoine Bruguier, Duc Le, Rohit Prabhavalkar et al.

We propose Neural-FST Class Language Model (NFCLM) for end-to-end speech recognition, a novel method that combines neural network language models (NNLMs) and finite state transducers (FSTs) in a mathematically consistent framework. Our method utilizes a background NNLM which models generic background text together with a collection of domain-specific entities modeled as individual FSTs. Each output token is generated by a mixture of these components; the mixture weights are estimated with a separately trained neural decider. We show that NFCLM significantly outperforms NNLM by 15.8% relative in terms of Word Error Rate. NFCLM achieves similar performance as traditional NNLM and FST shallow fusion while being less prone to overbiasing and 12 times more compact, making it more suitable for on-device usage.

Alex Xiao, Weiyi Zheng, Gil Keren et al.

With 4.5 million hours of English speech from 10 different sources across 120 countries and models of up to 10 billion parameters, we explore the frontiers of scale for automatic speech recognition. We propose data selection techniques to efficiently scale training data to find the most valuable samples in massive datasets. To efficiently scale model sizes, we leverage various optimizations such as sparse transducer loss and model sharding. By training 1-10B parameter universal English ASR models, we push the limits of speech recognition performance across many domains. Furthermore, our models learn powerful speech representations with zero and few-shot capabilities on novel domains and styles of speech, exceeding previous results across multiple in-house and public benchmarks. For speakers with disorders due to brain damage, our best zero-shot and few-shot models achieve 22% and 60% relative improvement on the AphasiaBank test set, respectively, while realizing the best performance on public social media videos. Furthermore, the same universal model reaches equivalent performance with 500x less in-domain data on the SPGISpeech financial-domain dataset.

Suyoun Kim, Duc Le, Weiyi Zheng et al.

Measuring automatic speech recognition (ASR) system quality is critical for creating user-satisfying voice-driven applications. Word Error Rate (WER) has been traditionally used to evaluate ASR system quality; however, it sometimes correlates poorly with user perception/judgement of transcription quality. This is because WER weighs every word equally and does not consider semantic correctness which has a higher impact on user perception. In this work, we propose evaluating ASR output hypotheses quality with SemDist that can measure semantic correctness by using the distance between the semantic vectors of the reference and hypothesis extracted from a pre-trained language model. Our experimental results of 71K and 36K user annotated ASR output quality show that SemDist achieves higher correlation with user perception than WER. We also show that SemDist has higher correlation with downstream Natural Language Understanding (NLU) tasks than WER.

Jay Mahadeokar, Yangyang Shi, Yuan Shangguan et al.

Often, the storage and computational constraints of embeddeddevices demand that a single on-device ASR model serve multiple use-cases / domains. In this paper, we propose aFlexibleTransducer(FlexiT) for on-device automatic speech recognition to flexibly deal with multiple use-cases / domains with different accuracy and latency requirements. Specifically, using a single compact model, FlexiT provides a fast response for voice commands, and accurate transcription but with more latency for dictation. In order to achieve flexible and better accuracy and latency trade-offs, the following techniques are used. Firstly, we propose using domain-specific altering of segment size for Emformer encoder that enables FlexiT to achieve flexible de-coding. Secondly, we use Alignment Restricted RNNT loss to achieve flexible fine-grained control on token emission latency for different domains. Finally, we add a domain indicator vector as an additional input to the FlexiT model. Using the combination of techniques, we show that a single model can be used to improve WERs and real time factor for dictation scenarios while maintaining optimal latency for voice commands use-cases

Yuan Shangguan, Rohit Prabhavalkar, Hang Su et al.

As speech-enabled devices such as smartphones and smart speakers become increasingly ubiquitous, there is growing interest in building automatic speech recognition (ASR) systems that can run directly on-device; end-to-end (E2E) speech recognition models such as recurrent neural network transducers and their variants have recently emerged as prime candidates for this task. Apart from being accurate and compact, such systems need to decode speech with low user-perceived latency (UPL), producing words as soon as they are spoken. This work examines the impact of various techniques - model architectures, training criteria, decoding hyperparameters, and endpointer parameters - on UPL. Our analyses suggest that measures of model size (parameters, input chunk sizes), or measures of computation (e.g., FLOPS, RTF) that reflect the model's ability to process input frames are not always strongly correlated with observed UPL. Thus, conventional algorithmic latency measurements might be inadequate in accurately capturing latency observed when models are deployed on embedded devices. Instead, we find that factors affecting token emission latency, and endpointing behavior have a larger impact on UPL. We achieve the best trade-off between latency and word error rate when performing ASR jointly with endpointing, while utilizing the recently proposed alignment regularization mechanism.

Duc Le, Mahaveer Jain, Gil Keren et al.

How to leverage dynamic contextual information in end-to-end speech recognition has remained an active research area. Previous solutions to this problem were either designed for specialized use cases that did not generalize well to open-domain scenarios, did not scale to large biasing lists, or underperformed on rare long-tail words. We address these limitations by proposing a novel solution that combines shallow fusion, trie-based deep biasing, and neural network language model contextualization. These techniques result in significant 19.5% relative Word Error Rate improvement over existing contextual biasing approaches and 5.4%-9.3% improvement compared to a strong hybrid baseline on both open-domain and constrained contextualization tasks, where the targets consist of mostly rare long-tail words. Our final system remains lightweight and modular, allowing for quick modification without model re-training.

Yangyang Shi, Varun Nagaraja, Chunyang Wu et al.

We propose a dynamic encoder transducer (DET) for on-device speech recognition. One DET model scales to multiple devices with different computation capacities without retraining or finetuning. To trading off accuracy and latency, DET assigns different encoders to decode different parts of an utterance. We apply and compare the layer dropout and the collaborative learning for DET training. The layer dropout method that randomly drops out encoder layers in the training phase, can do on-demand layer dropout in decoding. Collaborative learning jointly trains multiple encoders with different depths in one single model. Experiment results on Librispeech and in-house data show that DET provides a flexible accuracy and latency trade-off. Results on Librispeech show that the full-size encoder in DET relatively reduces the word error rate of the same size baseline by over 8%. The lightweight encoder in DET trained with collaborative learning reduces the model size by 25% but still gets similar WER as the full-size baseline. DET gets similar accuracy as a baseline model with better latency on a large in-house data set by assigning a lightweight encoder for the beginning part of one utterance and a full-size encoder for the rest.

Suyoun Kim, Abhinav Arora, Duc Le et al.

Word Error Rate (WER) has been the predominant metric used to evaluate the performance of automatic speech recognition (ASR) systems. However, WER is sometimes not a good indicator for downstream Natural Language Understanding (NLU) tasks, such as intent recognition, slot filling, and semantic parsing in task-oriented dialog systems. This is because WER takes into consideration only literal correctness instead of semantic correctness, the latter of which is typically more important for these downstream tasks. In this study, we propose a novel Semantic Distance (SemDist) measure as an alternative evaluation metric for ASR systems to address this issue. We define SemDist as the distance between a reference and hypothesis pair in a sentence-level embedding space. To represent the reference and hypothesis as a sentence embedding, we exploit RoBERTa, a state-of-the-art pre-trained deep contextualized language model based on the transformer architecture. We demonstrate the effectiveness of our proposed metric on various downstream tasks, including intent recognition, semantic parsing, and named entity recognition.

Duc Le, Gil Keren, Julian Chan et al.

End-to-end models in general, and Recurrent Neural Network Transducer (RNN-T) in particular, have gained significant traction in the automatic speech recognition community in the last few years due to their simplicity, compactness, and excellent performance on generic transcription tasks. However, these models are more challenging to personalize compared to traditional hybrid systems due to the lack of external language models and difficulties in recognizing rare long-tail words, specifically entity names. In this work, we present novel techniques to improve RNN-T's ability to model rare WordPieces, infuse extra information into the encoder, enable the use of alternative graphemic pronunciations, and perform deep fusion with personalized language models for more robust biasing. We show that these combined techniques result in 15.4%-34.5% relative Word Error Rate improvement compared to a strong RNN-T baseline which uses shallow fusion and text-to-speech augmentation. Our work helps push the boundary of RNN-T personalization and close the gap with hybrid systems on use cases where biasing and entity recognition are crucial.

Chunxi Liu, Frank Zhang, Duc Le et al.

End-to-end automatic speech recognition (ASR) models with a single neural network have recently demonstrated state-of-the-art results compared to conventional hybrid speech recognizers. Specifically, recurrent neural network transducer (RNN-T) has shown competitive ASR performance on various benchmarks. In this work, we examine ways in which RNN-T can achieve better ASR accuracy via performing auxiliary tasks. We propose (i) using the same auxiliary task as primary RNN-T ASR task, and (ii) performing context-dependent graphemic state prediction as in conventional hybrid modeling. In transcribing social media videos with varying training data size, we first evaluate the streaming ASR performance on three languages: Romanian, Turkish and German. We find that both proposed methods provide consistent improvements. Next, we observe that both auxiliary tasks demonstrate efficacy in learning deep transformer encoders for RNN-T criterion, thus achieving competitive results - 2.0%/4.2% WER on LibriSpeech test-clean/other - as compared to prior top performing models.

Jay Mahadeokar, Yuan Shangguan, Duc Le et al.

There is a growing interest in the speech community in developing Recurrent Neural Network Transducer (RNN-T) models for automatic speech recognition (ASR) applications. RNN-T is trained with a loss function that does not enforce temporal alignment of the training transcripts and audio. As a result, RNN-T models built with uni-directional long short term memory (LSTM) encoders tend to wait for longer spans of input audio, before streaming already decoded ASR tokens. In this work, we propose a modification to the RNN-T loss function and develop Alignment Restricted RNN-T (Ar-RNN-T) models, which utilize audio-text alignment information to guide the loss computation. We compare the proposed method with existing works, such as monotonic RNN-T, on LibriSpeech and in-house datasets. We show that the Ar-RNN-T loss provides a refined control to navigate the trade-offs between the token emission delays and the Word Error Rate (WER). The Ar-RNN-T models also improve downstream applications such as the ASR End-pointing by guaranteeing token emissions within any given range of latency. Moreover, the Ar-RNN-T loss allows for bigger batch sizes and 4 times higher throughput for our LSTM model architecture, enabling faster training and convergence on GPUs.

Suyoun Kim, Yuan Shangguan, Jay Mahadeokar et al.

Recurrent Neural Network Transducer (RNN-T), like most end-to-end speech recognition model architectures, has an implicit neural network language model (NNLM) and cannot easily leverage unpaired text data during training. Previous work has proposed various fusion methods to incorporate external NNLMs into end-to-end ASR to address this weakness. In this paper, we propose extensions to these techniques that allow RNN-T to exploit external NNLMs during both training and inference time, resulting in 13-18% relative Word Error Rate improvement on Librispeech compared to strong baselines. Furthermore, our methods do not incur extra algorithmic latency and allow for flexible plug-and-play of different NNLMs without re-training. We also share in-depth analysis to better understand the benefits of the different NNLM fusion methods. Our work provides a reliable technique for leveraging unpaired text data to significantly improve RNN-T while keeping the system streamable, flexible, and lightweight.

Yangyang Shi, Yongqiang Wang, Chunyang Wu et al.

This paper proposes an efficient memory transformer Emformer for low latency streaming speech recognition. In Emformer, the long-range history context is distilled into an augmented memory bank to reduce self-attention's computation complexity. A cache mechanism saves the computation for the key and value in self-attention for the left context. Emformer applies a parallelized block processing in training to support low latency models. We carry out experiments on benchmark LibriSpeech data. Under average latency of 960 ms, Emformer gets WER $2.50\%$ on test-clean and $5.62\%$ on test-other. Comparing with a strong baseline augmented memory transformer (AM-TRF), Emformer gets $4.6$ folds training speedup and $18\%$ relative real-time factor (RTF) reduction in decoding with relative WER reduction $17\%$ on test-clean and $9\%$ on test-other. For a low latency scenario with an average latency of 80 ms, Emformer achieves WER $3.01\%$ on test-clean and $7.09\%$ on test-other. Comparing with the LSTM baseline with the same latency and model size, Emformer gets relative WER reduction $9\%$ and $16\%$ on test-clean and test-other, respectively.

Matthew Perez, Wenyu Jin, Duc Le et al.

Speech is a critical biomarker for Huntington Disease (HD), with changes in speech increasing in severity as the disease progresses. Speech analyses are currently conducted using either transcriptions created manually by trained professionals or using global rating scales. Manual transcription is both expensive and time-consuming and global rating scales may lack sufficient sensitivity and fidelity. Ultimately, what is needed is an unobtrusive measure that can cheaply and continuously track disease progression. We present first steps towards the development of such a system, demonstrating the ability to automatically differentiate between healthy controls and individuals with HD using speech cues. The results provide evidence that objective analyses can be used to support clinical diagnoses, moving towards the tracking of symptomatology outside of laboratory and clinical environments.

Yangyang Shi, Yongqiang Wang, Chunyang Wu et al.

Transformers, originally proposed for natural language processing (NLP) tasks, have recently achieved great success in automatic speech recognition (ASR). However, adjacent acoustic units (i.e., frames) are highly correlated, and long-distance dependencies between them are weak, unlike text units. It suggests that ASR will likely benefit from sparse and localized attention. In this paper, we propose Weak-Attention Suppression (WAS), a method that dynamically induces sparsity in attention probabilities. We demonstrate that WAS leads to consistent Word Error Rate (WER) improvement over strong transformer baselines. On the widely used LibriSpeech benchmark, our proposed method reduced WER by 10%$ on test-clean and 5% on test-other for streamable transformers, resulting in a new state-of-the-art among streaming models. Further analysis shows that WAS learns to suppress attention of non-critical and redundant continuous acoustic frames, and is more likely to suppress past frames rather than future ones. It indicates the importance of lookahead in attention-based ASR models.

Ching-Feng Yeh, Jay Mahadeokar, Kaustubh Kalgaonkar et al.

We explore options to use Transformer networks in neural transducer for end-to-end speech recognition. Transformer networks use self-attention for sequence modeling and comes with advantages in parallel computation and capturing contexts. We propose 1) using VGGNet with causal convolution to incorporate positional information and reduce frame rate for efficient inference 2) using truncated self-attention to enable streaming for Transformer and reduce computational complexity. All experiments are conducted on the public LibriSpeech corpus. The proposed Transformer-Transducer outperforms neural transducer with LSTM/BLSTM networks and achieved word error rates of 6.37 % on the test-clean set and 15.30 % on the test-other set, while remaining streamable, compact with 45.7M parameters for the entire system, and computationally efficient with complexity of O(T), where T is input sequence length.

Duc Le, Thilo Koehler, Christian Fuegen et al.

Grapheme-based acoustic modeling has recently been shown to outperform phoneme-based approaches in both hybrid and end-to-end automatic speech recognition (ASR), even on non-phonemic languages like English. However, graphemic ASR still has problems with rare long-tail words that do not follow the standard spelling conventions seen in training, such as entity names. In this work, we present a novel method to train a statistical grapheme-to-grapheme (G2G) model on text-to-speech data that can rewrite an arbitrary character sequence into more phonetically consistent forms. We show that using G2G to provide alternative pronunciations during decoding reduces Word Error Rate by 3% to 11% relative over a strong graphemic baseline and bridges the gap on rare name recognition with an equivalent phonetic setup. Unlike many previously proposed methods, our method does not require any change to the acoustic model training procedure. This work reaffirms the efficacy of grapheme-based modeling and shows that specialized linguistic knowledge, when available, can be leveraged to improve graphemic ASR.

Yongqiang Wang, Abdelrahman Mohamed, Duc Le et al.

We propose and evaluate transformer-based acoustic models (AMs) for hybrid speech recognition. Several modeling choices are discussed in this work, including various positional embedding methods and an iterated loss to enable training deep transformers. We also present a preliminary study of using limited right context in transformer models, which makes it possible for streaming applications. We demonstrate that on the widely used Librispeech benchmark, our transformer-based AM outperforms the best published hybrid result by 19% to 26% relative when the standard n-gram language model (LM) is used. Combined with neural network LM for rescoring, our proposed approach achieves state-of-the-art results on Librispeech. Our findings are also confirmed on a much larger internal dataset.

Duc Le, Xiaohui Zhang, Weiyi Zheng et al.

There is an implicit assumption that traditional hybrid approaches for automatic speech recognition (ASR) cannot directly model graphemes and need to rely on phonetic lexicons to get competitive performance, especially on English which has poor grapheme-phoneme correspondence. In this work, we show for the first time that, on English, hybrid ASR systems can in fact model graphemes effectively by leveraging tied context-dependent graphemes, i.e., chenones. Our chenone-based systems significantly outperform equivalent senone baselines by 4.5% to 11.1% relative on three different English datasets. Our results on Librispeech are state-of-the-art compared to other hybrid approaches and competitive with previously published end-to-end numbers. Further analysis shows that chenones can better utilize powerful acoustic models and large training data, and require context- and position-dependent modeling to work well. Chenone-based systems also outperform senone baselines on proper noun and rare word recognition, an area where the latter is traditionally thought to have an advantage. Our work provides an alternative for end-to-end ASR and establishes that hybrid systems can be improved by dropping the reliance on phonetic knowledge.

Arman Shahbazian, Youn Kyu Lee, Duc Le et al.

Over the past three decades, considerable effort has been devoted to the study of software architecture. A major portion of this effort has focused on the originally proposed view of four "C"s---components, connectors, configurations, and constraints---that are the building blocks of a system's architecture. Despite being simple and appealing, this view has proven to be incomplete and has required further elaboration. To that end, researchers have more recently tried to approach architectures from another important perspective---that of design decisions that yield a system's architecture. These more recent efforts have lacked a precise understanding of several key questions, however: (1) What is an architectural design decision (definition)? (2) How can architectural design decisions be found in existing systems (identification)? (3) What system decisions are and are not architectural (classification)? (4) How are architectural design decisions manifested in the code (reification)? (5) How can important architectural decisions be preserved and/or changed as desired (evolution)? This paper presents a technique targeted at answering these questions by analyzing information that is readily available about software systems. We applied our technique on over 100 different versions of two widely adopted open- source systems, and found that it can accurately uncover the architectural design decisions embodied in the systems.