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

TL;DR

Teochew-Wild addresses the lack of publicly available Teochew resources by introducing an orthographically annotated in-the-wild speech corpus. It provides text-front-end tools for Grapheme-to-Phoneme conversion, polyphone disambiguation, and dialect vocabulary mapping, along with enhancements to the Teochew writing system. Experiments on TTS and ASR show the dataset supports intelligible synthesis (MOS up to 3.52) and competitive recognition performance, with large-model fine-tuning (Whisper) yielding strong results. This work establishes a foundational Teochew resource and accompanying tools to spur continued development of speech technologies for a prominent low-resource dialect.

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.

Figures (1)

• Figure 1: The Overview of Our Dataset Construction Workflow