Teochew Speech Dataset With Phonetic And Character Annotations

Teochew-Wild is a dataset of 12,500 audio clips from 20 native Teochew speakers. It was created by 'panlr' from online sources like news, storytelling, and TV programs, with annotations for standard characters and pinyin. The dataset was last updated in April 2026.

Use Cases

• Train automatic speech recognition models using the audio clips and their corresponding standard character annotations.
• Develop text-to-speech systems for Teochew by mapping the provided pinyin annotations to the audio waveforms.
• Study accent variation by analyzing audio from different regions like Chaozhou city, Shantou city, and Chenghai.
• Perform linguistic analysis on the mix of formal written and colloquial spoken language present in the corpus.

Strengths

• Contains 12,500 audio segments from 20 native speakers.
• Provides dual annotations for standard Chinese characters and Teochew pinyin.
• Covers multiple regional accents including Chaozhou, Shantou, and Chenghai.

Limitations

• The dataset size of 12,500 clips may be limited for training large-scale neural models.
• Audio sources are from public broadcasts and online media, which may introduce inherent recording quality variance.
• Specific metrics for annotation accuracy or audio quality are not provided.

Provenance

Source

Online audio-visual content including local news, storytelling programs, TV shows, and social media.

Collection Method

Data was collected, normalized, denoised, and segmented using an automated pipeline with manual corrections.

Time Range

null

Freshness

Last updated in April 2026.

Geography

Covers Teochew-speaking regions in Guangdong, China, including Chaozhou city, Shantou city, Chenghai, and Chao'an.