WenetSpeech-Chuan: A Large-Scale Sichuanese Corpus with Rich Annotation for Dialectal Speech Processing
Abstract
The scarcity of large-scale, open-source data for dialects severely hinders progress in speech technology, a challenge particularly acute for the widely spoken Sichuanese dialects of Chinese. To address this critical gap, we introduce WenetSpeech-Chuan, a 10,000-hour, richly annotated corpus constructed using our novel Chuan-Pipeline, a complete data processing framework for dialectal speech. To facilitate rigorous evaluation and demonstrate the corpus's effectiveness, we also release high-quality ASR and TTS benchmarks, WenetSpeech-Chuan-Eval, with manually verified transcriptions. Experiments show that models trained on WenetSpeech-Chuan achieve state-of-the-art performance among open-source systems and demonstrate results comparable to commercial services. As the largest open-source corpus for Sichuanese dialects, WenetSpeech-Chuan not only lowers the barrier to research in dialectal speech processing but also plays a crucial role in promoting AI equity and mitigating bias in speech technologies. The corpus, benchmarks, models, and receipts are publicly available on our project page.
The following analysis is generated by AI and may contain errors. It is separate from the authors’ abstract and is not peer review.
Research summary
The main contribution of this paper is the introduction of WenetSpeech-Chuan, a large-scale, richly annotated corpus for Sichuanese dialect speech processing, along with a comprehensive data processing framework and benchmarks that significantly enhance the field of dialectal speech technology. This work addresses a critical gap in available resources and sets a foundation for future advancements in speech recognition and synthesis for dialects.
Methodology
The paper presents a comprehensive methodology through the Chuan-Pipeline, which systematically processes and annotates a large-scale corpus of Sichuanese dialect speech. The multi-stage approach, including data acquisition, segmentation, quality assessment, and paralinguistic annotation, is well-structured and innovative. The integration of LLM-GER for transcription accuracy and the multi-modal punctuation prediction method are particularly noteworthy, showcasing a thoughtful application of existing technologies to address specific challenges in dialectal speech processing.
Results & evaluation
The experiments are robust, demonstrating the effectiveness of the WenetSpeech-Chuan corpus through rigorous evaluation against established benchmarks. The results indicate that models trained on this dataset achieve state-of-the-art performance in ASR and TTS tasks, with clear comparisons to both open-source and commercial systems. The detailed breakdown of performance metrics across various models provides valuable insights into the strengths of the proposed corpus.
Reproducibility
The paper emphasizes reproducibility by providing a detailed description of the Chuan-Pipeline and the datasets used. The release of the corpus, benchmarks, and models on a public GitHub repository enhances the potential for other researchers to replicate and build upon this work. However, the paper could benefit from more explicit instructions or documentation on the implementation of the pipeline.
Limitations
While the corpus is extensive, the paper does not address potential biases in the data collection process or the limitations of the models evaluated. Additionally, the focus on Sichuanese dialects may limit the generalizability of the findings to other dialects or languages. The reliance on existing ASR systems for transcription may also introduce errors that could affect the corpus's overall quality.
Broader impact
The WenetSpeech-Chuan corpus has significant implications for advancing research in dialectal speech processing, particularly for underrepresented dialects in AI technologies. By providing a large-scale, open-source resource, the authors contribute to promoting AI equity and reducing bias in speech technologies. This work can facilitate further research and development in ASR and TTS systems, ultimately benefiting a large population of Sichuanese speakers. The main contribution of this paper is the introduction of WenetSpeech-Chuan, a large-scale, richly annotated corpus for Sichuanese dialect speech processing, along with a comprehensive data processing framework and benchmarks that significantly enhance the field of dialectal speech technology. This work addresses a critical gap in available resources and sets a foundation for future advancements in speech recognition and synthesis for dialects.
Author of this paper? Send a correction.