Speech Communication最新文献

Traditionally, teaching a human and a Machine Learning (ML) model is quite different, but organized and structured learning has the ability to enable faster and better understanding of the underlying concepts. For example, when humans learn to speak, they first learn how to utter basic phones and then slowly move towards more complex structures such as words and sentences. Motivated by this observation, researchers have started to adapt this approach for training ML models. Since the main concept, the gradual increase in difficulty, resembles the notion of the curriculum in education, the methodology became known as Curriculum Learning (CL). In this work, we design and test new CL approaches to train Automatic Speech Recognition systems, specifically focusing on the so-called end-to-end models. These models consist of a single, large-scale neural network that performs the recognition task, in contrast to the traditional way of having several specialized components focusing on different subtasks (e.g., acoustic and language modeling). We demonstrate that end-to-end models can achieve better performances if they are provided with an organized training set consisting of examples that exhibit an increasing level of difficulty. To impose structure on the training set and to define the notion of an easy example, we explored multiple solutions that use either external, static scoring methods or incorporate feedback from the model itself. In addition, we examined the effect of pacing functions that control how much data is presented to the network during each training epoch. Our proposed curriculum learning strategies were tested on the task of speech recognition on two data sets, one containing spontaneous Finnish speech where volunteers were asked to speak about a given topic, and one containing planned English speech. Empirical results showed that a good curriculum strategy can yield performance improvements and speed-up convergence. After a given number of epochs, our best strategy achieved a 5.6% and 3.4% decrease in terms of test set word error rate for the Finnish and English data sets, respectively.

Recent research has shown evidence based on a minimal contrast paradigm that consonants and vowels are articulatorily synchronized at the onset of the syllable. What remains less clear is the laryngeal dimension of the syllable, for which evidence of tone synchrony with the consonant-vowel syllable has been circumstantial. The present study assesses the precise tone-vowel alignment in Mandarin Chinese by applying the minimal contrast paradigm. The vowel onset is determined by detecting divergence points of F2 trajectories between a pair of disyllabic sequences with two contrasting vowels, and the onsets of tones are determined by detecting divergence points of f₀ trajectories in contrasting disyllabic tone pairs, using generalized additive mixed models (GAMMs). The alignment of the divergence-determined vowel and tone onsets is then evaluated with linear mixed effect models (LMEMs) and their synchrony is validated with Bayes factors. The results indicate that tone and vowel onsets are fully synchronized. There is therefore evidence for strict alignment of consonant, vowel and tone as hypothesized in the synchronization model of the syllable. Also, with the newly established tone onset, the previously reported ‘anticipatory raising’ effect of tone now appears to occur within rather than before the articulatory syllable. Implications of these findings will be discussed.

This paper provides a structured examination of Arabic Automatic Speech Recognition (ASR), focusing on the complexity posed by the language’s diverse forms and dialectal variations. We first explore the Arabic language forms, delimiting the challenges encountered with Dialectal Arabic, including issues such as code-switching and non-standardized orthography and, thus, the scarcity of large annotated datasets. Subsequently, we delve into the landscape of Arabic resources, distinguishing between Modern Standard Arabic (MSA) and Dialectal Arabic (DA) Speech Resources and highlighting the disparities in available data between these two categories. Finally, we analyze both traditional and modern approaches in Arabic ASR, assessing their effectiveness in addressing the unique challenges inherent to the language. Through this comprehensive examination, we aim to provide insights into the current state and future directions of Arabic ASR research and development.

This study examined the vowel merger between the two vowels /e/ and /ɛ/ in Yanbian Korean. This sound change has already spread to Seoul Korean, particularly among speakers born after the 1970s. The aim of this study was to determine whether close exposure to Seoul Korean speakers leads to the neutralization of the distinction between the two vowels /e/ and /ɛ/. We recruited 20 Yanbian Korean speakers and asked them about their frequency of exposure to Seoul Korean. The exposure level of each participant was also recorded using a Likert scale. The results revealed that speakers with limited in-person interactions with Seoul Korean speakers exhibited distinct vowel productions within the vowel space. In contrast, those with frequent in-person interactions with Seoul Korean speakers tended to neutralize the two vowels, displaying considerably overlapping patterns in the vowel space. The relationship between the level of exposure to Seoul Korean and speakers’ vowel production was statistically confirmed by a linear regression analysis. Based on the results of this study, we speculate that the sound change in Yanbian Korean may become more widespread as Yanbian Korean speakers are increasingly exposed to Seoul Korean.

Although language delays are common in children with sex chromosome trisomies [SCT], no studies have analysed their prosodic abilities. Considering the importance of prosody in communication, this exploratory study aims to analyse the prosodic features of the narratives of 4-year-old children with SCT.

Participants included 22 children with SCT and 22 typically developing [TD] children. The Narrative Competence Task was administered to elicit the child's narrative. Each utterance was prosodically analysed considering pitch and timing variables.

Considering pitch, the only difference was the number of movements since the utterances of children with SCT were characterised by a lower speech modulation. However, considering the timing variables, children with SCT produced a faster speech rate and a shorter final syllable duration than TD children.

Since both speech modulation and duration measures have important syntactic and pragmatic functions, further investigations should deeply analyse the prosodic skills of children with SCT in interaction with syntax and pragmatics.

ECAPA-TDNN is one of the most popular TDNNs for speaker verification. While most of the updates pay attention to building precisely designed auxiliary modules, the depth-first principle has shown promising performance recently. However, empirical experiments show that one-dimensional convolution (Conv1D) based TDNNs suffer from performance degradation by simply adding massive vanilla basic blocks. Note that Conv1D naturally has a global receptive field (RF) on the feature dimension, progressive channel fusion (PCF) is proposed to alleviate this issue by introducing group convolution to build local RF and fusing the subbands progressively. Instead of reducing the group number in convolution layers used in the previous work, a novel channel permutation strategy is introduced to build information flow between groups so that all basic blocks in the model keep consistent parameter efficiency. The information leakage from lower-frequency bands to higher ones caused by Res2Block is simultaneously solved by introducing group-in-group convolution and using channel permutation. Besides the PCF strategy, redundant connections are removed for a more concise model architecture. The experiments on VoxCeleb and CnCeleb achieve state-of-the-art (SOTA) performance with an average relative improvement of 12.3% on EER and 13.2% on minDCF (0.01), validating the effectiveness of the proposed model.

Although end-to-end (E2E) trainable automatic speech recognition (ASR) has shown great success by jointly learning acoustic and linguistic information, it still suffers from the effect of domain shifts, thus limiting potential applications. The E2E ASR model implicitly learns an internal language model (LM) which characterises the training distribution of the source domain, and the E2E trainable nature makes the internal LM difficult to adapt to the target domain with text-only data. To solve this problem, this paper proposes decoupled structures for attention-based encoder–decoder (Decoupled-AED) and neural transducer (Decoupled-Transducer) models, which can achieve flexible domain adaptation in both offline and online scenarios while maintaining robust intra-domain performance. To this end, the acoustic and linguistic parts of the E2E model decoder (or prediction network) are decoupled, making the linguistic component (i.e. internal LM) replaceable. When encountering a domain shift, the internal LM can be directly replaced during inference by a target-domain LM, without re-training or using domain-specific paired speech-text data. Experiments for E2E ASR models trained on the LibriSpeech-100h corpus showed that the proposed decoupled structure gave 15.1% and 17.2% relative word error rate reductions on the TED-LIUM 2 and AESRC2020 corpora while still maintaining performance on intra-domain data. It is also shown that the decoupled structure can be used to boost cross-domain speech translation quality while retaining the intra-domain performance.

Speech-based automatic depression detection systems have been extensively explored over the past few years. Typically, each speaker is assigned a single label (Depressive or Non-depressive), and most approaches formulate depression detection as a speech classification task without explicitly considering the non-uniformly distributed depression pattern within segments, leading to low generalizability and robustness across different scenarios. However, depression corpora do not provide fine-grained labels (at the phoneme or word level) which makes the dynamic depression pattern in speech segments harder to track using conventional frameworks. To address this, we propose a novel framework, Speechformer-CTC, to model non-uniformly distributed depression characteristics within segments using a Connectionist Temporal Classification (CTC) objective function without the necessity of input–output alignment. Two novel CTC-label generation policies, namely the Expectation-One-Hot and the HuBERT policies, are proposed and incorporated in objectives on various granularities. Additionally, experiments using Automatic Speech Recognition (ASR) features are conducted to demonstrate the compatibility of the proposed method with content-based features. Our results show that the performance of depression detection, in terms of Macro F1-score, is improved on both DAIC-WOZ (English) and CONVERGE (Mandarin) datasets. On the DAIC-WOZ dataset, the system with HuBERT ASR features and a CTC objective optimized using HuBERT policy for label generation achieves 83.15% F1-score, which is close to state-of-the-art without the need for phoneme-level transcription or data augmentation. On the CONVERGE dataset, using Whisper features with the HuBERT policy improves the F1-score by 9.82% on CONVERGE1 (in-domain test set) and 18.47% on CONVERGE2 (out-of-domain test set). These findings show that depression detection can benefit from modeling non-uniformly distributed depression patterns and the proposed framework can be potentially used to determine significant depressive regions in speech utterances.

Trained on 680,000 h of massive speech data, Whisper is a multitasking, multilingual speech foundation model demonstrating superior performance in automatic speech recognition, translation, and language identification. However, its applicability in speaker verification (SV) tasks remains unexplored, particularly in low-data-resource scenarios where labeled speaker data in specific domains are limited. To fill this gap, we propose a lightweight adaptor framework to boost SV with Whisper, namely Whisper-SV. Given that Whisper is not specifically optimized for SV tasks, we introduce a representation selection module to quantify the speaker-specific characteristics contained in each layer of Whisper and select the top-k layers with prominent discriminative speaker features. To aggregate pivotal speaker-related features while diminishing non-speaker redundancies across the selected top-k distinct layers of Whisper, we design a multi-layer aggregation module in Whisper-SV to integrate multi-layer representations into a singular, compacted representation for SV. In the multi-layer aggregation module, we employ convolutional layers with shortcut connections among different layers to refine speaker characteristics derived from multi-layer representations from Whisper. In addition, an attention aggregation layer is used to reduce non-speaker interference and amplify speaker-specific cues for SV tasks. Finally, a simple classification module is used for speaker classification. Experiments on VoxCeleb1, FFSVC, and IMSV datasets demonstrate that Whisper-SV achieves EER/minDCF of 2.22%/0.307, 6.14%/0.488, and 7.50%/0.582, respectively, showing superior performance in low-data-resource SV scenarios.

Speaker modeling plays a crucial role in various tasks, and fixed-dimensional vector representations, known as speaker embeddings, are the predominant modeling approach. These embeddings are typically evaluated within the framework of speaker verification, yet their utility extends to a broad scope of related tasks including speaker diarization, speech synthesis, voice conversion, and target speaker extraction. This paper presents Wespeaker, a user-friendly toolkit designed for both research and production purposes, dedicated to the learning of speaker embeddings. Wespeaker offers scalable data management, state-of-the-art speaker embedding models, and self-supervised learning training schemes with the potential to leverage large-scale unlabeled real-world data. The toolkit incorporates structured recipes that have been successfully adopted in winning systems across various speaker verification challenges, ensuring highly competitive results. For production-oriented development, Wespeaker integrates CPU- and GPU-compatible deployment and runtime codes, supporting mainstream platforms such as Windows, Linux, Mac and on-device chips such as horizon X3’PI. Wespeaker also provides off-the-shelf high-quality speaker embeddings by providing various pretrained models, which can be effortlessly applied to different tasks that require speaker modeling. The toolkit is publicly available at https://github.com/wenet-e2e/wespeaker.