KeSpeech
1542-hour open-source Mandarin speech dataset covering 8 sub-dialects from 34 cities with 27,237 speakers
Duration
1542 hours
Languages
9
Sample Rate
16 kHz
Published
2021-12
Description
1Recorded by 27,237 speakers from 34 cities across China
2Includes standard Mandarin and 8 sub-dialects
3Provides multi-dimensional annotations: content transcription, speaker identity, and sub-dialect labels
4Supports multiple tasks including ASR, speaker verification, sub-dialect identification, and voice conversion
5Free for academic use
Language Details
| Language | Duration |
|---|---|
| Mandarin Chinese | None |
| Northeastern Mandarin | None |
| Central Plains Mandarin | None |
| Southwestern Mandarin | None |
| Jianghuai Mandarin | None |
| Wu Chinese | None |
| Cantonese | None |
| Southern Min | None |
| Hakka | None |
Publisher