Piper Zholi Zh Cn Checkpoints: Chinese TTS Model for Character Voice

A Chinese text-to-speech model trained to replicate the voice of the game character 'Zhongli'. The model was trained for approximately 48 hours over 150 epochs using about 45 hours of generated speech data cloned from the AISHELL-3 dataset. Created by LF83 and last updated on December 15, 2025, it is built on the Piper zh/zh_CN/huayan/medium foundation model.

Use Cases

• Generate Chinese speech in a specific character's voice for game dialogue based on the described voice cloning.
• Fine-tune TTS models for Mandarin Chinese using the provided checkpoints based on the Piper architecture.
• Study prosody and rhythm in synthesized Chinese speech based on the described text processing with inserted punctuation.
• Create audio content for media or entertainment featuring a consistent, cloned character voice.

Strengths

• Trained on approximately 45 hours of generated speech data, providing a substantial audio corpus.
• Uses the AISHELL-3 dataset as a source, which is a known resource for Chinese speech.
• Training configuration details are provided, including hardware (RTX 3090 GPU) and duration (48 hours).

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.

Provenance

Source

AISHELL-3 Chinese speech dataset, with voice cloning via IndexTTS2.

Collection Method

Text processed with FunASR for punctuation insertion, trained on RTX 3090 GPU.

Freshness

Last updated 2025-12-15 07:45:06; freshness should be verified.