piper-checkpoints-zh-AISHELL-3 (cn_aishell3)

Training checkpoint + config for the Mandarin voice cn_aishell3 (medium), trained on AISHELL-3 with Piper's pinyin frontend.

You usually do not need this repo. For inference, use the ready-to-run ONNX model in the companion repo:

https://huggingface.co/writeAPoemForYou/piper-voices-zh-AISHELL-3

Ready-to-run ONNX (recommended)

pip install piper-tts
pip install -U "huggingface_hub[cli]"

hf download writeAPoemForYou/piper-voices-zh-AISHELL-3 \
  zh/zh_CN/cn_aishell3/medium/zh_CN-cn_aishell3-medium.onnx \
  zh/zh_CN/cn_aishell3/medium/zh_CN-cn_aishell3-medium.onnx.json \
  zh/zh_CN/cn_aishell3/medium/tokens.txt \
  --local-dir cn_aishell3

# CLI
python -m piper --model cn_aishell3/zh_CN-cn_aishell3-medium.onnx \
  --speaker SSB0033 -f out.wav -i "这是一本书。"

# Python
from pathlib import Path
from piper import PiperVoice
from piper.config import SynthesisConfig

voice = PiperVoice.load(
    Path("cn_aishell3/zh_CN-cn_aishell3-medium.onnx"),
    download_dir=Path("cn_aishell3"),
)
for chunk in voice.synthesize("这是一本书。", syn_config=SynthesisConfig(speaker_id=0)):
    ...  # chunk.audio_float_array / chunk.sample_rate

Files in this repo

zh/zh_CN/cn_aishell3/medium/
  cn_aishell3-174.ckpt     # training checkpoint (Lightning), 174 speakers
  config.json              # Piper training config (phoneme_type=pinyin, PHONEME_TO_ID)
  MODEL_CARD

Export to ONNX yourself

pip install piper-tts
python -m piper.train.export_onnx \
  --checkpoint zh/zh_CN/cn_aishell3/medium/cn_aishell3-174.ckpt \
  --output-file zh_CN-cn_aishell3-medium.onnx

Dataset / License

• Dataset: AISHELL-3 — https://www.openslr.org/93/
• License: Apache License v.2.0
• Warm-started from en_US/libritts_r (medium), fine-tuned on AISHELL-3.

@inproceedings{AISHELL-3_2020,
  title={AISHELL-3: A Multi-speaker Mandarin TTS Corpus and the Baselines},
  author={Yao Shi, Hui Bu, Xin Xu, Shaoji Zhang, Ming Li},
  year={2015},
  url={https://arxiv.org/abs/2010.11567}
}