Accurately recognizing domain-specific words remains an arduous challenge facing current automatic speech recognition (ASR) systems. While there are a number of prior arts managing to incorporate domain context through crossattention mechanisms, these methods often add additional components that increase model complexity and focus solely on wordlevel cues, overlooking broader domain-level topic information. To address these limitations, we put forward PRIME, a simple yet effective prompt-tuning method that enriches ASR with domain context using LLM-generated topic descriptions. To mitigate the limited context window of the ASR decoder, we introduce a biasing word retriever that selects the most relevant domain-specific words to construct informative prompts. Notably, PRIME requires no architectural modifications and offers a lightweight, scalable solution for contextualized ASR. A series of experiments counducted on the AISHELL and SlideSpeech benchmark datasets show that PRIME considerably promotes biasing word recognition, outperforming some strong baselines.

Automatic pronunciation assessment (APA) manages to evaluate the pronunciation proficiency of a second language (L2) learner in a target language. Existing efforts typically draw on regression models for proficiency score prediction, wherein the models are trained to estimate target values without explicitly accounting for phoneme-awareness in the feature space. In this paper, we propose a contrastive phonemic ordinal regularizer (ConPCO) tailored for regression-based APA models to generate more phoneme-discriminative features while factoring in the ordinal relationships among the regression targets. The proposed ConPCO first aligns the phoneme representations of an APA model and textual embeddings of phonetic transcriptions via contrastive learning. Afterward, the phoneme characteristics are retained by regulating the distances between inter- and intra-phoneme categories in the feature space while allowing for the ordinal relationships among the output targets. We further design and develop a hierarchical APA model to evaluate the effectiveness of our regularizer. A series of experiments conducted on the speechocean762 benchmark dataset suggests the feasibility and effectiveness of our approach in relation to several competitive baselines.

Bi-Cheng Yan, Jiun-Ting Li, Yi-Cheng Wang, Hsin Wei Wang, Tien-Hong Lo, Yung-Chang Hsu, Wei-Cheng Chao, Berlin Chen. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

End-to-end automatic speech recognition (E2E ASR) systems often suffer from mistranscription of domain-specific phrases, such as named entities, sometimes leading to catastrophic failures in downstream tasks. A family of fast and lightweight named entity correction (NEC) models for ASR have recently been proposed, which normally build on pho-netic-level edit distance algorithms and have shown impressive NEC performance. However, as the named entity (NE) list grows, the problems of phonetic confusion in the NE list are exacerbated; for example, homophone ambiguities increase substantially. In view of this, we proposed a novel Description Augmented Named entity CorrEctoR (dubbed DANCER), which leverages entity descriptions to provide additional information to facilitate mitigation of phonetic con-fusion for NEC on ASR transcription. To this end, an efficient entity description augmented masked language model (EDA-MLM) comprised of a dense retrieval model is introduced, enabling MLM to adapt swiftly to domain-specific entities for the NEC task. A series of experiments conducted on the AISHELL-1 and Homophone datasets confirm the effectiveness of our modeling approach. DANCER outperforms a strong baseline, the phonetic edit-distance-based NEC model (PED-NEC), by a character error rate (CER) reduction of about 7% relatively on AISHELL-1 for named entities. More notably, when tested on Homophone that contain named entities of high phonetic confusion, DANCER offers a more pronounced CER reduction of 46% relatively over PED-NEC for named entities. The code is available at https://github.com/Amiannn/Dancer.

Automated speaking assessment in conversation tests (ASAC) aims to evaluate the overall speaking proficiency of an L2 (second-language) speaker in a setting where an interlocutor interacts with one or more candidates. Although prior ASAC approaches have shown promising performance on their respective datasets, there is still a dearth of research specifically focused on incorporating the coherence of the logical flow within a conversation into the grading model. To address this critical challenge, we propose a hierarchical graph model that aptly incorporates both broad inter-response interactions (e.g., discourse relations) and nuanced semantic information (e.g., semantic words and speaker intents), which is subsequently fused with contextual information for the final prediction. Extensive experimental results on the NICT-JLE benchmark dataset suggest that our proposed modeling approach can yield considerable improvements in prediction accuracy with respect to various assessment metrics, as compared to some strong baselines. This also sheds light on the importance of investigating coherence-related facets of spoken responses in ASAC.

End-to-end (E2E) automatic speech recognition (ASR) models have become standard practice for various commercial applications. However, in real-world scenarios, the long-tailed nature of word distribution often leads E2E ASR models to perform well on common words but fall short in recognizing uncommon ones. Recently, the notion of a contextual adapter (CA) was proposed to infuse external knowledge represented by a context word list into E2E ASR models. Although CA can improve recognition performance on rare words, two crucial data imbalance problems remain. First, when using low-frequency words as context words during training, since these words rarely occur in the utterance, CA becomes prone to overfit on attending to thetoken due to higher-frequency words not being present in the context list. Second, the long-tailed distribution within the context list itself still causes the model to perform poorly on low-frequency context words. In light of this, we explore in-depth the impact of altering the context list to have words with different frequency distributions on model performance, and meanwhile extend CA with a simple yet effective context-balanced learning objective1. A series of experiments conducted on the AISHELL-1 benchmark dataset suggests that using all vocabulary words from the training corpus as the context list and pairing them with our balanced objective yields the best performance, demonstrating a significant reduction in character error rate (CER) by up to 1.21% and a more pronounced 9.44% reduction in the error rate of zero-shot words.1The code is available at: https://github.com/Amiannn/espnet/tree/context_balanced_adapter

Automatic pronunciation assessment (APA) manages to quantify second language (L2) learners’ pronunciation proficiency in a target language by providing fine-grained feedback with multiple aspect scores (e.g., accuracy, fluency, and completeness) at various linguistic levels (i.e., phone, word, and utterance). Most of the existing efforts commonly follow a parallel modeling framework, which takes a sequence of phone-level pronunciation feature embeddings of a learner's utterance as input and then predicts multiple aspect scores across various linguistic levels. However, these approaches neither take the hierarchy of linguistic units into account nor consider the relatedness among the pronunciation aspects in an explicit manner. In light of this, we put forward an effective modeling approach for APA, termed HierGAT, which is grounded on a hierarchical graph attention network. Our approach facilitates hierarchical modeling of the input utterance as a heterogeneous graph that contains linguistic nodes at various levels of granularity. On top of the tactfully designed hierarchical graph message passing mechanism, intricate interdependencies within and across different linguistic levels are encapsulated and the language hierarchy of an utterance is factored in as well. Furthermore, we also design a novel aspect attention module to encode relatedness among aspects. To our knowledge, we are the first to introduce multiple types of linguistic nodes into graph-based neural networks for APA and perform a comprehensive qualitative analysis to investigate their merits. A series of experiments conducted on the speechocean762 benchmark dataset suggests the feasibility and effectiveness of our approach in relation to several competitive baselines.

Automatic pronunciation assessment (APA) manages to quantify the pronunciation proficiency of a second language (L2) learner in a language. Prevailing approaches to APA normally leverage neural models trained with a regression loss function, such as the mean-squared error (MSE) loss, for proficiency level prediction. Despite most regression models can effectively capture the ordinality of proficiency levels in the feature space, they are confronted with a primary obstacle that different phoneme categories with the same proficiency level are inevitably forced to be close to each other, retaining less phoneme-discriminative information. On account of this, we devise a phonemic contrast ordinal (PCO) loss for training regression-based APA models, which aims to preserve better phonemic distinctions between phoneme categories meanwhile considering ordinal relationships of the regression target output. Specifically, we introduce a phoneme-distinct regularizer into the MSE loss, which encourages feature representations of different phoneme categories to be far apart while simultaneously pulling closer the representations belonging to the same phoneme category by means of weighted distances. An extensive set of experiments carried out on the speechocean 762 benchmark dataset demonstrate the feasibility and effectiveness of our model in relation to some existing state-of-the-art models.

Mispronunciation detection and diagnosis (MDD) manages to pinpoint phone-level erroneous pronunciation segmentations and provide instant and informative diagnostic feedback to L2 (second-language) learners. Among the various modeling paradigms for MDD, dictation-based neural methods have recently become a de facto standard, which identifies pronunciation errors and returns diagnostic feedback at the same time by aligning the recognized phone sequence uttered by an L2 learner to the corresponding canonical phone sequence of a given text prompt. Despite their decent efficacy, dictation-based methods have at least two downsides. First, the dictation process and alignment process are made independent of each other, often resulting in a poor diagnostic feedback. Second, prior knowledge about the articulation traits of the canonical phones in the text prompt is not fully utilized in MDD. On account of this, we propose a novel end-to-end MDD method that can streamline the dictation process and the alignment process in a non-autoregressive manner. In addition, knowledge about phone-level articulation traits are extracted with a graph convolutional network (GCN) to obtain more discriminative phonetic embeddings so as to promote the MDD performance. An extensive set of experiments conducted on the L2-ARCTIC benchmark dataset suggest the feasibility and effectiveness of our approach in relation to competitive baselines.

End-to-end (E2E) neural modeling has emerged as one predominant school of thought to develop computer-assisted pronunciation training (CAPT) systems, showing competitive performance to conventional pronunciation-scoring based methods. However, current E2E neural methods for CAPT are faced with at least two pivotal challenges. On one hand, most of the E2E methods operate in an autoregressive manner with left-to-right beam search to dictate the pronunciations of an L2 learners. This however leads to very slow inference speed, which inevitably hinders their practical use. On the other hand, E2E neural methods are normally data-hungry and meanwhile an insufficient amount of nonnative training data would often reduce their efficacy on mispronunciation detection and diagnosis (MD&D). In response, we put forward a novel MD&D method that leverages non-autoregressive (NAR) E2E neural modeling to dramatically speed up the inference time while maintaining performance in line with the conventional E2E neural methods. In addition, we design and develop a pronunciation modeling network stacked on top of the NAR E2E models of our method to further boost the effectiveness of MD&D. Empirical experiments conducted on the L2-ARCTIC English dataset seems to validate the feasibility of our method, in comparison to some top-of-the-line E2E models and an iconic pronunciation-scoring based method built on a DNN-HMM acoustic model.

End-to-end (E2E) neural models are increasingly attracting attention as a promising modeling approach for mispronunciation detection and diagnosis (MDD). Typically, these models are trained by optimizing a cross-entropy criterion, which corresponds to improving the log-likelihood of the training data. However, there is a discrepancy between the objectives of model training and the MDD evaluation, since the performance of an MDD model is commonly evaluated in terms of F1-score instead of phone or word error rate (PER/WER). In view of this, we in this paper explore the use of a discriminative objective function for training E2E MDD models, which aims to maximize the expected F1-score directly. A series of experiments conducted on the L2-ARCTIC dataset show that our proposed method can yield considerable performance improvements in relation to some state-of-the-art E2E MDD approaches and the celebrated GOP method.

Conversational speech normally is embodied with loose syntactic structures at the utterance level but simultaneously exhibits topical coherence relations across consecutive utterances. Prior work has shown that capturing longer context information with a recurrent neural network or long short-term memory language model (LM) may suffer from the recent bias while excluding the long-range context. In order to capture the long-term semantic interactions among words and across utterances, we put forward disparate conversation history fusion methods for language modeling in automatic speech recognition (ASR) of conversational speech. Furthermore, a novel audio-fusion mechanism is introduced, which manages to fuse and utilize the acoustic embeddings of a current utterance and the semantic content of its corresponding conversation history in a cooperative way. To flesh out our ideas, we frame the ASR N-best hypothesis rescoring task as a prediction problem, leveraging BERT, an iconic pre-trained LM, as the ingredient vehicle to facilitate selection of the oracle hypothesis from a given N-best hypothesis list. Empirical experiments conducted on the AMI benchmark dataset seem to demonstrate the feasibility and efficacy of our methods in relation to some current top-of-line methods. The proposed methods not only achieve significant inference time reduction but also improve the ASR performance for conversational speech.

Mispronunciation detection and diagnosis (MDD) aims to detect erroneous pronunciation segments in an L2 learner's articulation and subsequently provide informative diagnostic feedback. Most existing neural methods follow a dictation-based modeling paradigm that finds out pronunciation errors and returns diagnostic feedback at the same time by aligning the recognized phone sequence uttered by an L2 learner to the corresponding canonical phone sequence of a given text prompt. However, the main downside of these methods is that the dictation process and alignment process are mostly made independent of each other. In view of this, we present a novel end-to-end neural method, dubbed PeppaNet, building on a unified structure that can jointly model the dictation process and the alignment process. The model of our method learns to directly predict the pronunciation correctness of each canonical phone of the text prompt and in turn provides its corresponding diagnostic feedback. In contrast to the conventional dictation-based methods that rely mainly on a free-phone recognition process, PeppaNet makes good use of an effective selective gating mechanism to simultaneously incorporate phonetic, phonological and acoustic cues to generate corrections that are more proper and phonetically related to the canonical pronunciations. Extensive sets of experiments conducted on the L2-ARCTIC benchmark dataset seem to show the merits of our proposed method in comparison to some recent top-of-the-line methods.

Due to the unprecedented breakthroughs brought about by deep learning, speech enhancement (SE) techniques have been developed rapidly and play an important role prior to acoustic modeling so as to mitigate noise effects on speech. To increase the perceptual quality of speech, the current state-of-the-art in the realm of SE adopts adversarial training by connecting an objective metric to the discriminator. However, there is no guarantee that optimizing the perceptual quality of speech will necessarily lead to improved automatic speech recognition (ASR) performance. In this study, we present TENET††Inspired by the movie - TENET, Christopher Nolan, 2020., **Some of the enhanced audio samples can be found from https://fuann.github.io/TENET., a novel Time-reversal Enhancement NETwork, which leverages the transformation of an input noisy signal itself, i.e., the time-reversed version, in conjunction with a Siamese network and a complex dual-path Transformer to promote SE performance for noise-robust ASR. Extensive experiments conducted on the Voicebank-DEMAND dataset show that TENET can achieve stellar results compared to a few top-of-the-line methods in terms of both SE and ASR evaluation metrics. To demonstrate the model generalization ability, we further evaluate TENET on the test set of scenarios contaminated with unseen noise, and the results also confirm the superiority of this promising method.

With the acceleration of globalization, more and more people are willing or required to learn second languages (L2). One of the major remaining challenges facing current mispronunciation and diagnosis (MDD) models for use in computer-assisted pronunciation training (CAPT) is to handle speech from L2 learners with a diverse set of accents. In this paper, we set out to mitigate the adverse effects of accent variety in building an L2 English MDD system with end-to-end (E2E) neural models. To this end, we first propose an effective modeling framework that infuses accent features into an E2E MDD model, thereby making the model more accent-aware. Going a step further, we design and present disparate accent-aware modules to perform accent-aware modulation of acoustic features in a finer-grained manner, so as to enhance the discriminating capability of the resulting MDD model. Extensive sets of experiments conducted on the L2-ARCTIC benchmark dataset show the merits of our MDD model, in comparison to some existing E2E-based strong baselines and the celebrated pronunciation scoring based method.

Mispronunciation detection and diagnosis (MDD) is a core component of computer-assisted pronunciation training (CAPT).Most of the existing MDD approaches focus on dealing with categorical errors (viz.one canonical phone is substituted by another one, aside from those mispronunciations caused by deletions or insertions).However, accurate detection and diagnosis of non-categorial or distortion errors (viz.approximating L2 phones with L1 (first-language) phones, or erroneous pronunciations in between) still seems out of reach.In view of this, we propose to conduct MDD with a novel endto-end automatic speech recognition (E2E-based ASR) approach.In particular, we expand the original L2 phone set with their corresponding anti-phone set, making the E2E-based MDD approach have a better capability to take in both categorical and non-categorial mispronunciations, aiming to provide better mispronunciation detection and diagnosis feedback.Furthermore, a novel transfer-learning paradigm is devised to obtain the initial model estimate of the E2E-based MDD system without resource to any phonological rules.Extensive sets of experimental results on the L2-ARCTIC dataset show that our best system can outperform the existing E2E baseline system and pronunciation scoring based method (GOP) in terms of the F1-score, by 11.05% and 27.71%, respectively.Index

Noise robustness has long garnered much interest from researchers and practitioners of the automatic speech recognition (ASR) community due to its paramount importance to the success of ASR systems. This paper presents a novel approach to improving the noise robustness of speech features, building on top of the dictionary learning paradigm. To this end, we employ the K-SVD method and its variants to create sparse representations with respect to a common set of basis spectral vectors that captures the intrinsic temporal structure inherent in the modulation spectra of clean training speech features. The enhanced modulation spectra of speech features, constructed by mapping the original modulation spectra into the space spanned by these representative basis vectors, can better carry noise-resistant acoustic characteristics. In addition, considering the nonnegative property of the modulation spectrum amplitudes, we utilize the nonnegative K-SVD method, in combination with the nonnegative sparse coding method, to generate more noise-robust speech features. All experiments were conducted and verified using the standard Aurora-2 database and task. The empirical results show that the proposed dictionary learning based approach can provide significant average word error reductions when being integrated with either a GMM-HMM or a DNN-HMM based ASR system.