Zhen Wan, Chao-Han Huck Yang, Jinchuan Tian, Hanrong Ye, Ankita Pasad, Szu-Wei Fu, Arushi Goel, Ryo Hachiuma, Shizhe Diao, Kunal Dhawan, Sreyan Ghosh, Yusuke Hirota, Zhehuai Chen, Rafael Valle, Chenhui Chu, Shinji Watanabe, Boris Ginsburg, Yu-Chiang Frank Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Zhen Wan, Chao-Han Huck Yang, Yahan Yu, Jinchuan Tian, Sheng Li, Ke Hu, Zhehuai Chen, Shinji Watanabe, Fei Cheng, Chenhui Chu, Sadao Kurohashi. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

We present the system demonstration and opensource code release of a novel, data-efficient framework that converts any standard text Large Language Model (LLM) into a full-duplex end-to-end (E2E) speech-to-speech (S2S) model, for building conversational voice agents. Our new modeling method enables any LLMs to simultaneously listen and speak without requiring extensive speech-text pretraining. Moreover, we demonstrate how to put together a low-latency and full-duplex voice agent with open-source modeling, inference optimization, and serving solutions. This work significantly lowers the barrier to entry for developing low-latency, human-like voice agents by providing a generalizable, end-to-end solution built on open-source technologies.

Large language models (LLMs) have demonstrated remarkable advancements in language understanding and generation. Building on the success of text-based LLMs, recent research has adapted these models to use speech embeddings for prompting, resulting in Speech-LLM models that exhibit strong performance in automatic speech recognition (ASR) and automatic speech translation (AST). In this work, we propose a novel approach to leverage ASR transcripts as prompts for AST in a Speech-LLM built on an encoder-decoder text LLM. The Speech-LLM model consists of a speech encoder and an encoder-decoder structure Megatron-T5. By first decoding speech to generate ASR transcripts and subsequently using these transcripts along with encoded speech for prompting, we guide the speech translation in a two-step process like chain-of-thought (CoT) prompting. Low-rank adaptation (LoRA) is used for the T5 LLM for model adaptation and shows superior performance to full model fine-tuning. Experimental results show that the proposed CoT prompting significantly improves AST performance, achieving an average increase of 2.4 BLEU points across 6 En→X or X→En AST tasks compared to speech prompting alone. Additionally, compared to a related CoT prediction method that predicts a concatenated sequence of ASR and AST transcripts, our method performs better by an average of 2 BLEU points.

Recent end-to-end speech language models (SLMs) have expanded upon the capabilities of large language models (LLMs) by incorporating pre-trained speech models. However, these SLMs often undergo extensive speech instruction-tuning to bridge the gap between speech and text modalities. This requires significant annotation efforts and risks catastrophic forgetting of the original language capabilities. In this work, we present a simple yet effective automatic process for creating speech-text pair data that carefully injects speech paralinguistic understanding abilities into SLMs while preserving the inherent language capabilities of the text-based LLM. Our model demonstrates general capabilities for speech-related tasks without the need for speech instruction-tuning data, achieving impressive performance on Dynamic-SUPERB and AIR-Bench-Chat benchmarks. Furthermore, our model exhibits the ability to follow complex instructions derived from LLMs, such as specific output formatting and chain-of-thought reasoning. Our approach not only enhances the versatility and effectiveness of SLMs but also reduces reliance on extensive annotated datasets, paving the way for more efficient and capable speech understanding systems.1

A rising interest in the modality extension of foundation language models warrants discussion on the most effective, and efficient, multimodal training approach. This work focuses on neural machine translation (NMT) and proposes a joint multimodal training regime of Speech-LLM to include automatic speech translation (AST). We investigate two different foundation model architectures, decoder-only GPT and encoder-decoder T5, extended with Canary-1B’s speech encoder. To handle joint multimodal training, we propose a novel training framework called EMMeTT. EMMeTT improves training efficiency with the following: balanced sampling across languages, datasets, and modalities; efficient sequential data iteration; and a novel 2D bucketing scheme for multimodal data, complemented by a batch size optimizer (OOMptimizer). We show that a multimodal training consistently helps with both architectures. Moreover, SALM-T5 trained with EMMeTT retains the original NMT capability while outperforming AST baselines on four-language subsets of FLORES and FLEURS. The resultant Multimodal Translation Model produces strong text and speech translation results at the same time.

An ideal multimodal agent should be aware of the quality of its input modalities. Recent advances have enabled large language models (LLMs) to incorporate auditory systems for handling various speech-related tasks. However, most audio LLMs remain unaware of the quality of the speech they process. This limitation arises because speech quality evaluation is typically excluded from multi-task training due to the lack of suitable datasets. To address this, we introduce the first natural language-based speech evaluation corpus, generated from authentic human ratings. In addition to the overall Mean Opinion Score (MOS), this corpus offers detailed analysis across multiple dimensions and identifies causes of quality degradation. It also enables descriptive comparisons between two speech samples (A/B tests) with human-like judgment. Leveraging this corpus, we propose an alignment approach with LLM distillation (ALLD) to guide the audio LLM in extracting relevant information from raw speech and generating meaningful responses. Experimental results demonstrate that ALLD outperforms the previous state-of-the-art regression model in MOS prediction, with a mean square error of 0.17 and an A/B test accuracy of 98.6%. Additionally, the generated responses achieve BLEU scores of 25.8 and 30.2 on two tasks, surpassing the capabilities of task-specific models. This work advances the comprehensive perception of speech signals by audio LLMs, contributing to the development of real-world auditory and sensory intelligent agents.

Siqi Ouyang, Oleksii Hrinchuk, Zhehuai Chen, Vitaly Lavrukhin, Jagadeesh Balam, Lei Li, Boris Ginsburg. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Yifan Peng, Krishna C Puvvada, Zhehuai Chen, Piotr Zelasko, He Huang, Kunal Dhawan, Ke Hu, Shinji Watanabe, Jagadeesh Balam, Boris Ginsburg. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Yuchen Hu, Chen Chen, Chao-Han Huck Yang, Ruizhe Li, Dong Zhang, Zhehuai Chen, Eng Siong Chng. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

We present a novel Speech Augmented Language Model (SALM) with multitask and in-context learning capabilities. SALM comprises a frozen text LLM, a audio encoder, a modality adapter module, and LoRA layers to accommodate speech input and associated task instructions. The unified SALM not only achieves performance on par with task-specific Conformer baselines for Automatic Speech Recognition (ASR) and Speech Translation (AST), but also exhibits zero-shot in-context learning capabilities, demonstrated through keyword-boosting task for ASR and AST. Moreover, speech supervised in-context training is proposed to bridge the gap between LLM training and downstream speech tasks, which further boosts the in-context learning ability of speech-to-text models. Proposed model is open-sourced via NeMo toolkit1.

This paper proposes Transducers with Pronunciation-aware Embeddings (PET). Unlike conventional Transducers where the decoder embeddings for different tokens are trained independently, the PET model’s decoder embedding incorporates shared components for text tokens with the same or similar pronunciations. With experiments conducted in multiple datasets in Mandarin Chinese and Korean, we show that PET models consistently improve speech recognition accuracy compared to conventional Transducers. Our investigation also uncovers a phenomenon that we call error chain reactions. Instead of recognition errors being evenly spread throughout an utterance, they tend to group together, with subsequent errors often following earlier ones. Our analysis shows that PET models effectively mitigate this issue by substantially reducing the likelihood of the model generating additional errors following a prior one. Our implementation will be open-sourced with the NeMo toolkit.

Incorporating speech understanding capabilities into pretrained large-language models has become a vital research direction (SpeechLLM). The previous architectures can be categorized as: i) GPT-style, prepend speech prompts to the text prompts as a sequence of LLM inputs like a decoder-only model; ii) T5-style, introduce speech cross-attention to each layer of the pretrained LLMs. We propose BESTOW architecture to bring the BESt features from $T w O$ Worlds into a single model that is highly efficient and has strong multitask capabilities. Moreover, there is no clear streaming solution for either style, especially considering the solution should generalize to speech multitask. We reformulate streamable SpeechLLM as a read-write policy problem and unifies the offline and streaming research with BESTOW architecture. Hence we demonstrate the first open-source SpeechLLM solution that enables Streaming and Multitask at scale (beyond ASR) at the same time. This streamable solution achieves very strong performance on a wide range of speech tasks (ASR, AST, SQA, unseen DynamicSuperb). It is end-to-end optimizable, with lower training/inference cost, and demonstrates LLM knowledge transferability to speech.

Neural speech editing advancements have raised concerns about their misuse in spoofing attacks. Traditional partially edited speech corpora primarily focus on cut-and-paste edits, which, while maintaining speaker consistency, often introduce detectable discontinuities. Recent methods, like $\mathrm{A}^{3} \mathrm{~T}$ and Voicebox, improve transitions by leveraging contextual information. To foster spoofing detection research, we introduce the Speech INfilling Edit (SINE) dataset, created with Voicebox. We detailed the process of re-implementing Voicebox training and dataset creation. Subjective evaluations confirm that speech edited using this novel technique is more challenging to detect than conventional cut-and-paste methods. Despite human difficulty, experimental results demonstrate that self-supervised-based detectors can achieve remarkable performance in detection, localization, and generalization across different edit methods. The dataset and related models will be made available at: https://jasonswfu.github.io/SINE_dataset/index.html

Given recent advances in generative AI technology, a key question is how large language models (LLMs) can enhance acoustic modeling tasks using text decoding results from a frozen, pretrained automatic speech recognition (ASR) model. To explore new capabilities in language modeling for speech processing, we introduce the generative speech transcription error correction (GenSEC) challenge. This challenge comprises three post-ASR language modeling tasks: (i) post-ASR transcription correction, (ii) speaker tagging, and (iii) emotion recognition. These tasks aim to emulate future LLM-based agents handling voice-based interfaces while remaining accessible to a broad audience by utilizing open pretrained language models or agent-based APIs. We also discuss insights from baseline evaluations, as well as lessons learned for designing future evaluations.

This paper proposes Virtuoso, a massively multilingual speech–text joint semi-supervised learning framework for text-to-speech synthesis (TTS) models. Existing multilingual TTS typically supports tens of languages, which are a small fraction of the thousands of languages in the world. One difficulty to scale multilingual TTS to hundreds of languages is collecting high-quality speech–text paired data in low-resource languages. This study extends Maestro, a speech–text joint pretraining framework for automatic speech recognition (ASR), to speech generation tasks. To train a TTS model from various types of speech and text data, different training schemes are designed to handle supervised (paired TTS and ASR data) and unsupervised (untranscribed speech and unspoken text) datasets. Experimental evaluation shows that 1) multilingual TTS models trained on Virtuoso can achieve significantly better naturalness and intelligibility than baseline ones in seen languages, and 2) they can synthesize reasonably intelligible and naturally sounding speech for unseen languages where no high-quality paired TTS data is available.

We propose a novel method to accelerate training and inference process of recurrent neural network transducer (RNN-T) based on the guidance from a co-trained connectionist temporal classification (CTC) model. We made a key assumption that if an encoder embedding frame is classified as a blank frame by the CTC model, it is likely that this frame will be aligned to blank for all the partial alignments or hypotheses in RNN-T and it can be discarded from the decoder input. We also show that this frame reduction operation can be applied in the middle of the encoder, which result in significant speed up for the training and inference in RNN-T. We further show that the CTC alignment, a by-product of the CTC decoder, can also be used to perform lattice reduction for RNN-T during training. Our method is evaluated on the Librispeech and SpeechStew tasks. We demonstrate that the proposed method is able to accelerate the RNN-T inference by 2.2 times with similar or slightly better word error rates (WER).

Recently, a number of approaches to train speech models by incorporating text into end-to-end models have been developed, with Maestro advancing state-of-the-art automatic speech recognition (ASR) and Speech Translation (ST) performance. In this paper, we expand our understanding of the resulting shared speech-text representations with two types of analyses. First we examine the limits of speech-free domain adaptation, finding that a corpus-specific duration model for speech-text alignment is the most important component for learning a shared speech-text representation. Second, we inspect the similarities between activations of unimodal (speech or text) encoders as compared to the activations of a shared encoder. We find that the shared encoder learns a more compact and overlapping speech-text representation than the uni-modal encoders. We hypothesize that this partially explains the effectiveness of the Maestro shared speech-text representations.

An effective way to learn representations from untranscribed speech and unspoken text with linguistic/lexical representations derived from synthesized speech was introduced in tts4pretrain [1]. However, the representations learned from synthesized and real speech are likely to be different, potentially limiting the improvements from incorporating unspoken text. In this paper, we introduce learning from supervised speech earlier on in the training process with consistency-based regularization between real and synthesized speech. This allows for better learning of shared speech and text representations. Thus, we introduce a new objective, with encoder and decoder consistency and contrastive regularization between real and synthesized speech derived from the labeled corpora during the pretraining stage. We show that the new objective leads to more similar representations derived from speech and text that help downstream ASR. The proposed pretraining method yields Word Error Rate (WER) reductions of 7-21% relative on six public corpora, Librispeech, AMI, TEDLIUM, Common Voice, Switchboard, CHiME-6, over a state-of-the-art baseline pretrained with wav2vec2.0 and 2-17% over the previously proposed tts4pretrain. The proposed method outperforms the supervised SpeechStew by up to 17%. Moreover, we show that the proposed method also yields WER reductions on larger data sets by evaluating on a large resource, in-house Voice Search task and streaming ASR.

We present Maestro, a self-supervised training method to unify representations learnt from speech and text modalities.Self-supervised learning from speech signals aims to learn the latent structure inherent in the signal, while self-supervised learning from text attempts to capture lexical information.Learning aligned representations from unpaired speech and text sequences is a challenging task.Previous work either implicitly enforced the representations learnt from these two modalities to be aligned in the latent space through multitasking and parameter sharing or explicitly through conversion of modalities via speech synthesis.While the former suffers from interference between the two modalities, the latter introduces additional complexity.In this paper, we propose Maestro, a novel algorithm to learn unified representations from both these modalities simultaneously that can transfer to diverse downstream tasks such as Automated Speech Recognition (ASR) and Speech Translation (ST).Maestro learns unified representations through sequence alignment, duration prediction and matching embeddings in the learned space through an aligned masked-language model loss.We establish a new state-of-the-art (SOTA) on VoxPopuli multilingual ASR with a 8% relative reduction in Word Error Rate (WER), multidomain SpeechStew ASR (3.7% relative) and 21 languages to English multilingual ST on CoVoST 2 with an improvement of 2.8 BLEU averaged over 21 languages.

Self-supervised learning of speech representations has achieved impressive results in improving automatic speech recognition (ASR).In this paper, we show that data selection is important for self-supervised learning.We propose a simple and effective unsupervised data selection method which selects acoustically similar speech to a target domain.It takes the discrete speech representation available in common self-supervised learning frameworks as input, and applies a contrastive data selection method on the discrete tokens.Through extensive empirical studies we show that our proposed method reduces the amount of required pre-training data and improves the downstream ASR performance.Pre-training on a selected subset of 6% of the general data pool results in 11.8% relative improvements in LibriSpeech test-other compared to pre-training on the full set.On Multilingual LibriSpeech French, German, and Spanish test sets, selecting 6% data for pre-training reduces word error rate by more than 15% relatively compared to the full set, and achieves competitive results compared to current state-of-theart performances.

Training state-of-the-art Automated Speech Recognition (ASR) models typically requires a substantial amount of transcribed speech. In this work, we demonstrate that a modality-matched joint speech and text model introduced in [1] can be leveraged to train a massively multilingual ASR model without any supervised (manually transcribed) speech for some languages. This paper explores the use of jointly learnt speech and text representations in a massively multilingual, zero supervised speech, real-world setting to expand the set of languages covered by ASR with only unlabeled speech and text in the target languages. Using the FLEURS dataset, we define the task to cover 102 languages, where transcribed speech is available in 52 of these languages and can be used to improve end-to-end ASR quality on the remaining 50. First, we show that by combining speech representations with byte-level text representations and use of language embeddings, we can dramatically reduce the Character Error Rate (CER) on languages with no supervised speech from 64.8% to 30.8%, a relative reduction of 53%. Second, using a subset of South Asian languages we show that Maestro-U can promote knowledge transfer from languages with supervised speech even when there is limited to no graphemic overlap. Overall, Maestro-U closes the gap to oracle performance by 68.5% relative and reduces the CER of 19 languages below 15%.

We present JOIST, an algorithm to train a streaming, cascaded, encoder end-to-end (E2E) model with both speech-text paired inputs, and text-only unpaired inputs. Unlike previous works, we explore joint training with both modalities, rather than pre-training and fine-tuning. In addition, we explore JOIST using a streaming E2E model with an order of magnitude more data, which are also novelties compared to previous works. Through a series of ablation studies, we explore different types of text modeling, including how to model the length of the text sequence and the appropriate text subword unit representation. We find that best text representation for JOIST improves WER across a variety of search and rare-word test sets by 4-14% relative, compared to a model not trained with text. In addition, we quantitatively show that JOIST maintains streaming capabilities, which is important for good user-level experience.

Self-supervised pretraining for Automated Speech Recognition (ASR) has shown varied degrees of success. In this paper, we propose to jointly learn representations during pretraining from two different modalities: speech and text. The proposed method, tts4pretrain complements the power of contrastive learning in self-supervision with linguistic/lexical representations derived from synthesized speech, effectively learning from untranscribed speech and unspoken text. Lexical learning in the speech encoder is enforced through an additional sequence loss term that is coupled with contrastive loss during pretraining. We demonstrate that this novel pretraining method yields Word Error Rate (WER) reductions of 10% relative on the well-benchmarked, Librispeech task over a state-of-the-art baseline pretrained with wav2vec2.0 only. The proposed method also serves as an effective strategy to compensate for the lack of transcribed speech, effectively matching the performance of 5000 hours of transcribed speech with just 100 hours of transcribed speech on the AMI meeting transcription task. Finally, we demonstrate WER reductions of up to 15% on an inhouse Voice Search task over traditional pretraining. Incorporating text into encoder pretraining is complimentary to rescoring with a larger or in-domain language model, resulting in additional 6% relative reduction in WER.

We introduce asynchronous dynamic decoder, which adopts an efficient A* algorithm to incorporate big language models in the one-pass decoding for large vocabulary continuous speech recognition. Unlike standard one-pass decoding with on-the-fly composition decoder which might induce a significant computation overhead, the asynchronous dynamic decoder has a novel design where it has two fronts, with one performing "exploration" and the other "backfill". The computation of the two fronts alternates in the decoding process, resulting in more effective pruning than the standard one-pass decoding with an on-the-fly composition decoder. Experiments show that the proposed decoder works notably faster than the standard one-pass decoding with on-the-fly composition decoder, while the acceleration will be more obvious with the increment of data complexity.

Speech synthesis has advanced to the point of being close to indistinguishable from human speech. However, efforts to train speech recognition systems on synthesized utterances have not been able to show that synthesized data can be effectively used to augment or replace human speech. In this work, we demonstrate that promoting consistent predictions in response to real and synthesized speech enables significantly improved speech recognition performance. We also find that training on 460 hours of LibriSpeech augmented with 500 hours of transcripts (without audio) performance is within 0.2% WER of a system trained on 960 hours of transcribed audio. This suggests that with this approach, when there is sufficient text available, reliance on transcribed audio can be cut nearly in half.

End-to-end (E2E) systems have played a more and more important role in automatic speech recognition (ASR) and achieved great performance. However, E2E systems recognize output word sequences directly with the input acoustic feature, which can only be trained on limited acoustic data. The extra text data is widely used to improve the results of traditional artificial neural network-hidden Markov model (ANN-HMM) hybrid systems. The involving of extra text data to standard E2E ASR systems may break the E2E property during decoding. In this paper, a novel modular E2E ASR system is proposed. The modular E2E ASR system consists of two parts: an acoustic-to-phoneme (A2P) model and a phoneme-to-word (P2W) model. The A2P model is trained on acoustic data, while extra data including large scale text data can be used to train the P2W model. This additional data enables the modular E2E ASR system to model not only the acoustic part but also the language part. During the decoding phase, the two models will be integrated and act as a standard acoustic-to-word (A2W) model. In other words, the proposed modular E2E ASR system can be easily trained with extra text data and decoded in the same way as a standard E2E ASR system. Experimental results on the Switchboard corpus show that the modular E2E model achieves better word error rate (WER) than standard A2W models.

We introduce a lattice determinization algorithm that can operate incrementally. That is, a word-level lattice can be generated for a partial utterance and then, once we have processed more audio, we can obtain a word-level lattice for the extended utterance without redoing all the work of lattice determinization. This is relevant for ASR decoders such as those used in Kaldi, which first generate a state-level lattice and then convert it to a word-level lattice using a determinization algorithm in a special semiring. Our incremental determinization algorithm is useful when word-level lattices are needed prior to the end of the utterance, and also reduces the latency due to determinization at the end of the utterance.

End-to-end modeling (E2E) of automatic speech recognition (ASR) blends all the components of a traditional speech recognition system into a single, unified model. Although it simplifies the ASR systems, the unified model is hard to adapt when training and testing data mismatches. In this work, we focus on contextual speech recognition, which is particularly challenging for E2E models because contextual information is only available in inference time. To improve the performance in the presence of contextual information during training, we propose to use class-based language models (CLM) that can populate context-dependent information during inference. To enable this approach to scale to a large number of class members and minimize search errors, we propose a token passing algorithm with an efficient token recombination for E2E systems. We evaluate the proposed system on general and contextual ASR tasks, and achieve relative 62% Word Error Rate (WER) reduction for the contextual ASR task without hurting recognition performance for the general ASR task. We also show that the proposed method performs well without modification of the decoding hyper-parameters across tasks, making it a desirable solution for E2E ASR.

Unsupervised single-channel overlapped speech recognition is one of the hardest problems in automatic speech recognition (ASR). The problems can be modularized into three sub-problems: frame-wise interpreting, sequence level speaker tracing and speech recognition. Nevertheless, previous acoustic models formulate the correlation between sequential labels implicitly, which limit the modeling effect. In this work, we include explicit models for the sequential label correlation during training. This is relevant to models given by both the feature sequence and the output of the last frame. Moreover, we propose to integrate the linguistic information into the assignment decision of the permutation invariant training (PIT). Namely, a senone level neural network language model (NNLM) trained in the clean speech alignment is integrated, while the objective function is still cross-entropy. The proposed methods can be combined with an improved version of PIT and sequence discriminative training, which brings about further over 10% relative improvement of WER in the artificial overlapped Switchboard and hub5e-swb dataset.

End-to-end (E2E) automatic speech recognition (ASR) systems directly map acoustics to words using a unified model. Previous works mostly focus on E2E training a single model which integrates acoustic and language model into a whole. Although E2E training benefits from sequence modeling and simplified decoding pipelines, large amount of transcribed acoustic data is usually required, and traditional acoustic and language modelling techniques cannot be utilized. In this paper, a novel modular training framework of E2E ASR is proposed to separately train neural acoustic and language models during training stage, while still performing end-to-end inference in decoding stage. Here, an acoustics-to-phoneme model (A2P) and a phoneme-to-word model (P2W) are trained using acoustic data and text data respectively. A phone synchronous decoding (PSD) module is inserted between A2P and P2W to reduce sequence lengths without precision loss. Finally, modules are integrated into an acoustics-to-word model (A2W) and jointly optimized using acoustic data to retain the advantage of sequence modeling. Experiments on a 300-hour Switchboard task show significant improvement over the direct A2W model. The efficiency in both training and decoding also benefits from the proposed method.

We describe initial work on an extension of the Kaldi toolkit that supports weighted finite-state transducer (WFST) decoding on Graphics Processing Units (GPUs). We implement token recombination as an atomic GPU operation in order to fully parallelize the Viterbi beam search, and propose a dynamic load balancing strategy for more efficient token passing scheduling among GPU threads. We also redesign the exact lattice generation and lattice pruning algorithms for better utilization of the GPUs. Experiments on the Switchboard corpus show that the proposed method achieves identical 1-best results and lattice quality in recognition and confidence measure tasks, while running 3 to 15 times faster than the single process Kaldi decoder. The above results are reported on different GPU architectures. Additionally we obtain a 46-fold speedup with sequence parallelism and multi-process service (MPS) in GPU.

Unsupervised single-channel overlapped speech recognition is one of the hardest problems in automatic speech recognition (ASR). Permutation invariant training (PIT) is a state of the art model-based approach, which applies a single neural network to solve this single-input, multiple-output modeling problem. We propose to advance the current state of the art by imposing a modular structure on the neural network, applying a progressive pretraining regimen, and improving the objective function with transfer learning and a discriminative training criterion. The modular structure splits the problem into three subtasks: frame-wise interpreting, utterance-level speaker tracing, and speech recognition. The pretraining regimen uses these modules to solve progressively harder tasks. Transfer learning leverages parallel clean speech to improve the training targets for the network. Our discriminative training formulation is a modification of standard formulations that also penalizes competing outputs of the system. Experiments are conducted on the artificial overlapped switchboard and hub5e-swb dataset. The proposed framework achieves over 30% relative improvement of word error rate over both a strong jointly trained system, PIT for ASR, and a separately optimized system, PIT for speech separation with clean speech ASR model. The improvement comes from better model generalization, training efficiency, and the sequence level linguistic knowledge integration.

Connectionist Temporal Classification (CTC) model has achieved state-of-the-art LVCSR performance. However, due to the introduction of the blank symbol, word-level confidence measures (CM) based on CTC model can not be easily calculated by directly using the traditional phone posterior normalization or confusion network (CN) approaches. Recently, a phone synchronous decoding (PSD) framework has been proposed for efficient decoding with CTC model. By automatically ignoring blank frames, PSD decoding not only achieves significant speed-up, but also yields highly compact and precise CTC phone lattices. In this work, two CM generation approaches on top of the PSD CTC lattice are proposed. Detailed investigation is also carried out to demonstrate the effectiveness of PSD CTC lattice. Experiments on an English switchboard LVCSR task showed that the performance of the proposed PSD CTC lattice based CM can significantly outperform the CM based on traditional frame synchronous decoding with CTC or HMM models.

Connectionist temporal classification (CTC) has recently shown improved performance and efficiency in automatic speech recognition. One popular decoding implementation is to use a CTC model to predict the phone posteriors at each frame and then perform Viterbi beam search on a modified WFST network. This is still within the traditional frame synchronous decoding framework. In this paper, the peaky posterior property of CTC is carefully investigated and it is found that ignoring blank frames will not introduce additional search errors. Based on this phenomenon, a novel phone synchronous decoding framework is proposed by removing tremendous search redundancy due to blank frames, which results in significant search speed up. The framework naturally leads to an extremely compact phone-level acoustic space representation: CTC lattice. With CTC lattice, efficient and effective modular speech recognition approaches, second pass rescoring for large vocabulary continuous speech recognition (LVCSR), and phone-based keyword spotting (KWS), are also proposed in this paper. Experiments showed that phone synchronous decoding can achieve 3-4 times search speed up without performance degradation compared to frame synchronous decoding. Modular LVCSR with CTC lattice can achieve further WER improvement. KWS with CTC lattice not only achieved significant equal error rate improvement, but also greatly reduced the KWS model size and increased the search speed.