Joonas Kalda , Séverin Baroudi , Martin Lebourdais , Clément Pagés , Ricard Marxer , Tanel Alumäe , Hervé Bredin
{"title":"Design choices for PixIT-based speaker-attributed ASR: Team ToTaTo at the NOTSOFAR-1 challenge","authors":"Joonas Kalda , Séverin Baroudi , Martin Lebourdais , Clément Pagés , Ricard Marxer , Tanel Alumäe , Hervé Bredin","doi":"10.1016/j.csl.2025.101824","DOIUrl":"10.1016/j.csl.2025.101824","url":null,"abstract":"<div><div>PixIT is a recently proposed joint training framework that integrates Permutation Invariant Training (PIT) for speaker diarization and Mixture Invariant Training (MixIT) for speech separation. By leveraging diarization labels, PixIT addresses MixIT’s limitations, producing aligned sources and speaker activations that enable automatic long-form separation. We investigate applications of PixIT on the speaker-attributed automatic speech recognition (SA-ASR) task based on our systems for the NOTSOFAR-1 Challenge. We explore modifications to the joint ToTaToNet by integrating advanced self-supervised learning (SSL) features and masking networks. We show that fine-tuning an ASR system on PixIT-separated sources significantly boosts downstream SA-ASR performance, outperforming standard diarization-based baselines without relying on synthetic data. We explore lightweight post-processing heuristics for improving SA-ASR timestamp errors caused by long silences and artifacts present in file-level separated sources. We also show the potential of extracting speaker embeddings for the diarization pipeline directly from separated sources, with performance rivaling standard methods without any fine-tuning of speaker embeddings. On the NOTSOFAR-1 Challenge dataset, our PixIT-based approach outperforms the CSS-based baseline by 20% in terms of tcpWER after fine-tuning the ASR system on the separated sources. Notably, even when using the same ASR model as the baseline, our system is able to outperform it, without using any of the provided domain-specific synthetic data. These advancements position PixIT as a robust and flexible solution for real-world SA-ASR.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101824"},"PeriodicalIF":3.1,"publicationDate":"2025-05-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144131279","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Towards decoupling frontend enhancement and backend recognition in monaural robust ASR","authors":"Yufeng Yang , Ashutosh Pandey , DeLiang Wang","doi":"10.1016/j.csl.2025.101821","DOIUrl":"10.1016/j.csl.2025.101821","url":null,"abstract":"<div><div>It has been shown that the intelligibility of noisy speech can be improved by speech enhancement (SE) algorithms. However, monaural SE has not been established as an effective frontend for automatic speech recognition (ASR) in noisy conditions compared to an ASR model trained on noisy speech directly. The divide between SE and ASR impedes the progress of robust ASR systems, especially as SE has made major advances in recent years. This paper focuses on eliminating this divide with an ARN (attentive recurrent network) time-domain, a TF-CrossNet time–frequency domain, and an MP-SENet magnitude-phase based enhancement model. The proposed systems decouple frontend enhancement and backend ASR, with the latter trained only on clean speech. Results on the WSJ, CHiME-2, LibriSpeech, and CHiME-4 corpora demonstrate that ARN, TF-CrossNet, and MP-SENet enhanced speech all translate to improved ASR results in noisy and reverberant environments, and generalize well to real acoustic scenarios. The proposed system outperforms the baselines trained on corrupted speech directly. Furthermore, it cuts the previous best word error rate (WER) on CHiME-2 by 28.4% relatively with a 5.6% WER, and achieves <span><math><mrow><mn>3</mn><mo>.</mo><mn>3</mn><mo>/</mo><mn>4</mn><mo>.</mo><mn>4</mn><mtext>%</mtext></mrow></math></span> WER on single-channel CHiME-4 simulated/real test data without training on CHiME-4. We also observe consistent improvements using noise-robust Whisper as the backend ASR model.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101821"},"PeriodicalIF":3.1,"publicationDate":"2025-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144106632","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"An item response theory framework to evaluate automatic speech recognition systems against speech difficulty","authors":"Chaina Santos Oliveira, Ricardo B.C. Prudêncio","doi":"10.1016/j.csl.2025.101817","DOIUrl":"10.1016/j.csl.2025.101817","url":null,"abstract":"<div><div>Evaluating the performance of Automatic Speech Recognition (ASR) systems is very relevant for selecting good techniques and understanding their advantages and limitations. ASR systems are usually evaluated by adopting test sets of audio speeches, ideally with different difficulty levels. In this sense, it is important to analyse whether a system under test correctly transcribes easy test speeches, while being robust to the most difficult ones. In this paper, a novel framework is proposed for evaluating ASR systems, which covers two complementary issues: (1) to measure the difficulty of each test speech; and (2) to analyse each ASR system’s performance against the difficulty level. Regarding the first issue, the framework measures speech difficulty by adopting Item Response Theory (IRT). Regarding the second issue, the Recognizer Characteristic Curve (RCC) is proposed, which is a plot of the ASR system’s performance versus speech difficulty. ASR performance is further analysed by a two-dimensional plot, in which speech difficulty is decomposed by IRT into sentence difficulty and speaker quality. In the experiments, the proposed framework was applied in a test set produced by adopting text-to-speech tools, with diverse speakers and sentences. Additionally, noise injection was applied to produce test items with even higher difficulty levels. In the experiments, noise injection actually increases difficulty and generates a wide variety of speeches to assess ASR performance. However, it is essential to pay attention that high noise levels can lead to an unreliable evaluation. The proposed plots were helpful for both identifying robust ASR systems as well as for choosing the noise level that results in both diversity and reliability.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101817"},"PeriodicalIF":3.1,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144072019","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Paige Tuttösí , Mantaj Dhillon , Luna Sang , Shane Eastwood , Poorvi Bhatia , Quang Minh Dinh , Avni Kapoor , Yewon Jin , Angelica Lim
{"title":"BERSting at the screams: A benchmark for distanced, emotional and shouted speech recognition","authors":"Paige Tuttösí , Mantaj Dhillon , Luna Sang , Shane Eastwood , Poorvi Bhatia , Quang Minh Dinh , Avni Kapoor , Yewon Jin , Angelica Lim","doi":"10.1016/j.csl.2025.101815","DOIUrl":"10.1016/j.csl.2025.101815","url":null,"abstract":"<div><div>Some speech recognition tasks, such as automatic speech recognition (ASR), are approaching or have reached human performance in many reported metrics. Yet, they continue to struggle in complex, real-world, situations, such as with distanced speech. Previous challenges have released datasets to address the issue of distanced ASR, however, the focus remains primarily on distance, specifically relying on multi-microphone array systems. Here we present the B(asic) E(motion) R(andom phrase) S(hou)t(s) (BERSt) dataset. The dataset contains almost 4 h of English speech from 98 actors with varying regional and non-native accents. The data was collected on smartphones in the actors homes and therefore includes at least 98 different acoustic environments. The data also includes 7 different emotion prompts and both shouted and spoken utterances. The smartphones were places in 19 different positions, including obstructions and being in a different room than the actor. This data is publicly available for use and can be used to evaluate a variety of speech recognition tasks, including: ASR, shout detection, and speech emotion recognition (SER). We provide initial benchmarks for ASR and SER tasks, and find that ASR degrades both with an increase in distance and shout level and shows varied performance depending on the intended emotion. Our results show that the BERSt dataset is challenging for both ASR and SER tasks and continued work is needed to improve the robustness of such systems for more accurate real-world use.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101815"},"PeriodicalIF":3.1,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144106633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Fan Yang , Tan Zhu , Jing Huang , Zhilin Huang , Guoqi Xie
{"title":"A novel graph kernel algorithm for improving the effect of text classification","authors":"Fan Yang , Tan Zhu , Jing Huang , Zhilin Huang , Guoqi Xie","doi":"10.1016/j.csl.2025.101818","DOIUrl":"10.1016/j.csl.2025.101818","url":null,"abstract":"<div><div>Text classification is an important topic in natural language processing. In recent years, both graph kernel methods and deep learning methods have been widely employed in text classification tasks. However, previous graph kernel algorithms focused too much on the graph structure itself, such as the shortest path subgraph,while focusing limited attention to the information of the text itself. Previous deep learning methods have often resulted in substantial utilization of computational resources. Therefore,we propose a new graph kernel algorithm to address the disadvantages. First,we extract the textual information of the document using the term weighting scheme. Second,we collect the structural information on the document graph. Third, graph kernel is used for similarity measurement for text classification.</div><div>We compared eight baseline methods on three experimental datasets, including traditional deep learning methods and graph-based classification methods, and tested our algorithm on multiple indicators. The experimental results demonstrate that our algorithm outperforms other baseline methods in terms of accuracy. Furthermore, it achieves a minimum reduction of 69% in memory consumption and a minimum decrease of 23% in runtime. Furthermore, as we decrease the percentage of training data, our algorithm continues to achieve superior results compared to other deep learning methods. The excellent experimental results show that our algorithm can improve the efficiency of text classification tasks and reduce the occupation of computer resources under the premise of ensuring high accuracy.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101818"},"PeriodicalIF":3.1,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144106631","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Qinwen Hu , Tianchi Sun , Xin’an Chen , Xiaobin Rong , Jing Lu
{"title":"Optimization of modular multi-speaker distant conversational speech recognition","authors":"Qinwen Hu , Tianchi Sun , Xin’an Chen , Xiaobin Rong , Jing Lu","doi":"10.1016/j.csl.2025.101816","DOIUrl":"10.1016/j.csl.2025.101816","url":null,"abstract":"<div><div>Conducting multi-speaker distant conversational speech recognition on real meeting recordings is a challenging task and has recently become an active area of research. In this work, we focus on modular approaches to addressing this challenge, integrating continuous speech separation (CSS), automatic speech recognition (ASR), and speaker diarization in a pipeline. We explore the effective utilization of the high-performing separation model, TF-GridNet, within our system and propose integration techniques to enhance the performance of the ASR and diarization modules. Our system is evaluated on both the LibriCSS and the real-world CHiME-8 NOTSOFAR-1 dataset. Through a comprehensive analysis of the system’s generalization performance, we identify key areas for further improvement in the front-end module.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101816"},"PeriodicalIF":3.1,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144106629","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"An end-to-end integration of speech separation and recognition with self-supervised learning representation","authors":"Yoshiki Masuyama , Xuankai Chang , Wangyou Zhang , Samuele Cornell , Zhong-Qiu Wang , Nobutaka Ono , Yanmin Qian , Shinji Watanabe","doi":"10.1016/j.csl.2025.101813","DOIUrl":"10.1016/j.csl.2025.101813","url":null,"abstract":"<div><div>Multi-speaker automatic speech recognition (ASR) has gained growing attention in a wide range of applications, including conversation analysis and human–computer interaction. Speech separation and enhancement (SSE) and single-speaker ASR have witnessed remarkable performance improvements with the rapid advances in deep learning. Complex spectral mapping predicts the short-time Fourier transform (STFT) coefficients of each speaker and has achieved promising results in several SSE benchmarks. Meanwhile, self-supervised learning representation (SSLR) has demonstrated its significant advantage in single-speaker ASR. In this work, we push forward the performance of multi-speaker ASR under noisy reverberant conditions by integrating powerful SSE, SSL, and ASR models in an end-to-end manner. We systematically investigate both monaural and multi-channel SSE methods and various feature representations. Our experiments demonstrate the advantages of recently proposed complex spectral mapping and SSLRs in multi-speaker ASR. The experimental results also confirm that end-to-end fine-tuning with an ASR criterion is important to achieve state-of-the-art word error rates (WERs) even with powerful pre-trained models. Moreover, we show the performance trade-off between SSE and ASE and mitigate it with a multi-task learning framework with both SSE and ASR criteria.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101813"},"PeriodicalIF":3.1,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143948781","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Francisco Teixeira , Karla Pizzi , Raphaël Olivier , Alberto Abad , Bhiksha Raj , Isabel Trancoso
{"title":"Exploring features for membership inference in ASR model auditing","authors":"Francisco Teixeira , Karla Pizzi , Raphaël Olivier , Alberto Abad , Bhiksha Raj , Isabel Trancoso","doi":"10.1016/j.csl.2025.101812","DOIUrl":"10.1016/j.csl.2025.101812","url":null,"abstract":"<div><div>Membership inference (MI) poses a substantial privacy threat to the training data of automatic speech recognition (ASR) systems, while also offering an opportunity to audit these models with regard to user data. This paper explores the effectiveness of loss-based features in combination with Gaussian and adversarial perturbations to perform MI in ASR models. We compare our proposed features with commonly used error-based features for both sample-level and speaker-level MI. We find that the proposed features greatly enhance performance for sample-level MI. For speaker-level MI, these features improve results, though by a smaller margin, as error-based features already obtain a high performance for this task. Our findings emphasise the importance of considering different feature sets and levels of access to target models for effective MI in ASR systems, providing valuable insights for auditing such models.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101812"},"PeriodicalIF":3.1,"publicationDate":"2025-05-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144072018","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Modality fusion using auxiliary tasks for dementia detection","authors":"Hangshou Shao, Yilin Pan, Yue Wang, Yijia Zhang","doi":"10.1016/j.csl.2025.101814","DOIUrl":"10.1016/j.csl.2025.101814","url":null,"abstract":"<div><div>Alzheimer’s disease is the leading cause of dementia that affects elderly individual’s speech and language abilities. In this paper, a <strong>F</strong>eature <strong>F</strong>usion Model with <strong>G</strong>uide Patterns (FFG) is designed as an acoustic- and linguistic-based dementia detection system, considering the limited publicly available data and modalities fusion inefficiency. Specifically, a multi-modal features interaction module composed of multiple co-attention layers is designed to improve multi-modal interaction between the acoustic and linguistic information embedded in the audio recordings. Given the limited audio recordings available in public datasets, guide patterns are introduced as auxiliary tasks to enhance the interaction between acoustic and linguistic information. Our proposed FFG model is evaluated with three publicly available datasets, namely, Pitt, ADReSS, and ADReSSo. Experimental results demonstrate that the FFG model can achieve superior resu lts on all three publicly available datasets. An exceptional performance of 85.85% and 84.30% accuracy was achieved on the Pitt and ADReSSo datasets. The ablation study demonstrated the efficiency of our proposed model.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"95 ","pages":"Article 101814"},"PeriodicalIF":3.1,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143942714","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Combined generative and predictive modeling for speech super-resolution","authors":"Heming Wang , Eric W. Healy , DeLiang Wang","doi":"10.1016/j.csl.2025.101808","DOIUrl":"10.1016/j.csl.2025.101808","url":null,"abstract":"<div><div>Speech super-resolution (SR) is the task that restores high-resolution speech from low-resolution input. Existing models employ simulated data and constrained experimental settings, which limit generalization to real-world SR. Predictive models are known to perform well in fixed experimental settings, but can introduce artifacts in adverse conditions. On the other hand, generative models learn the distribution of target data and have a better capacity to perform well on unseen conditions. In this study, we propose a novel two-stage approach that combines the strengths of predictive and generative models. Specifically, we employ a diffusion-based model that is conditioned on the output of a predictive model. Our experiments demonstrate that the model significantly outperforms single-stage counterparts and existing strong baselines on benchmark SR datasets. Furthermore, we introduce a repainting technique during the inference of the diffusion process, enabling the proposed model to regenerate high-frequency components even in mismatched conditions. An additional contribution is the collection of and evaluation on real SR recordings, using the same microphone at different native sampling rates. We make this dataset freely accessible, to accelerate progress towards real-world speech super-resolution.</div></div>","PeriodicalId":50638,"journal":{"name":"Computer Speech and Language","volume":"94 ","pages":"Article 101808"},"PeriodicalIF":3.1,"publicationDate":"2025-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143929481","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}