Teochew-Wild: The First In-the-wild Teochew Dataset with Orthographic Annotations

2025-05-08

AI Summary

To address the scarcity of high-quality spontaneous speech data and standardized annotations for the low-resource Teochew dialect, this work introduces the first Teochew speech dataset featuring precise character-level orthographic transcriptions and phonemic (Teochew Romanization Scheme, “ChaoPin”) annotations. The dataset comprises 18.9 hours of multi-speaker, multi-scenario spontaneous speech, covering both formal and colloquial registers. We establish a systematic annotation protocol and develop an integrated toolkit for audio cleaning, orthographic normalization, ChaoPin transcription, and text preprocessing. Empirical evaluation on automatic speech recognition (ASR) and text-to-speech (TTS) tasks demonstrates substantial improvements in modeling the speech–text mapping for Teochew. Both the dataset and toolchain are fully open-sourced, bridging critical gaps in spontaneous speech resources and unified annotation standards for Teochew—and more broadly, offering a reusable foundation for low-resource dialect speech research.

Technology Category

• Natural Language Processing: Speech
• Machine Learning: Large Multimodal Models (LMMs)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Web Mining and Content Analysis: Web data integration and cleaning

Abstract

This paper reports the construction of the Teochew-Wild, a speech corpus of the Teochew dialect. The corpus includes 18.9 hours of in-the-wild Teochew speech data from multiple speakers, covering both formal and colloquial expressions, with precise orthographic and pinyin annotations. Additionally, we provide supplementary text processing tools and resources to propel research and applications in speech tasks for this low-resource language, such as automatic speech recognition (ASR) and text-to-speech (TTS). To the best of our knowledge, this is the first publicly available Teochew dataset with accurate orthographic annotations. We conduct experiments on the corpus, and the results validate its effectiveness in ASR and TTS tasks.

Problem

• Constructing first annotated Teochew speech dataset
• Providing tools for low-resource language research
• Validating dataset effectiveness in ASR and TTS

Innovation

• First public Teochew dataset with orthographic annotations
• Includes 18.9 hours of diverse in-the-wild speech
• Provides supplementary tools for ASR and TTS