---
license: apache-2.0
language:
  - zh
task_categories:
  - text-to-speech
pretty_name: aishell3-ready (AISHELL-3 Mandarin, 24 kHz parquet)
size_categories:
  - 10K<n<100K
tags:
  - speech
  - tts
  - mandarin
  - aishell-3
---

# aishell3-ready

Mandarin Chinese multi-speaker speech from **AISHELL-3** (Beijing Shell Shell Technology Co., Ltd., Apache License 2.0), repackaged as 24 kHz FLAC-in-parquet shards for TTS and aligner training.

| Item | Value |
|---|---|
| Source | AISHELL-3 (44.1 kHz, 218 native Mandarin speakers, ~85 h) |
| Rows | 88,035 utterances |
| Audio | 24 kHz mono FLAC bytes in `audio` (resampled from 44.1 kHz, even-length trimmed) |
| Shards | 45 × `part-NNNNN.parquet` (2,000 rows each) |
| Columns | `uid`, `audio`, `num_samples`, `sampling_rate`, `text`, `phonemes`, `speaker`, `speaker_id`, `language`, `split` |
| Sidecars | `_v9/` phoneme sidecars (IPA symbol strings with Mandarin tone marks, stress, word index, character spans), joined on `uid` |

`speaker` keeps the AISHELL-3 speaker code (`aishell3:SSB0005`, …). A companion repository `dabsdamoon/aishell3-ready-f0` adds pre-extracted CREPE pitch columns to the same rows.

## License and attribution

AISHELL-3 is distributed under the **Apache License 2.0**. Please cite:

- Shi, Y., Bu, H., Xu, X., Zhang, S., Li, M. *AISHELL-3: A Multi-speaker Mandarin TTS Corpus and the Baselines* (2020). https://www.openslr.org/93/

Our repackaging (resampling, parquet layout, phoneme sidecars) is released under the same Apache 2.0 terms.