chattts_local

本地 ChatTTS 语音合成技能。使用 ChatTTS 模型将文字转换为自然流畅的中文语音。完全本地运行，免费无需 API Key。支持调节语速、音调、情感。使用场景：(1) QQ 消息语音回复 (2) 文档朗读 (3) 通知提醒语音化 (4) 长文本转语音

Security Scan

The skill's files and instructions are consistent with a local ChatTTS text-to-speech tool, but it fetches model files from an unverified mirror and auto-installs pip packages—review those endpoints and package sources before running.

Purpose & Capability

Name/description, SKILL.md, and included scripts all implement a local ChatTTS TTS pipeline: a tts.py runner, model-download helpers, and instructions to pip-install ChatTTS/torch/torchaudio. There are no unrelated credentials, binaries, or capabilities requested.

Instruction Scope

SKILL.md instructs running scripts/tts.py and installing dependencies; the scripts auto-install pip packages if missing, set HF_ENDPOINT to a mirror, check/create ~/.openclaw/ChatTTS/models, and download or load models. Behavior stays within TTS scope, but the automatic pip install and automatic model download are actions with side effects the user should be aware of.

Install Mechanism

No formal install spec (instruction-only) but provided scripts download model artifacts. The downloads point to https://hf-mirror.com (both via wget in download_models.sh and by setting HF_ENDPOINT in Python). This is a third‑party mirror (not huggingface.co) and is an external network dependency; users should verify/trust the mirror. download_models.sh and download_models.py will write ~500MB into ~/.openclaw/ChatTTS/models and snapshot_download is invoked with force_download=True.

Credentials

The skill declares no required env vars or credentials. The code sets HF_ENDPOINT internally (so no secret is required) and does not access unrelated system credentials or config paths. No excessive env/credential access is requested.

Persistence & Privilege

always is false and the skill does not modify other skills or global agent settings. It creates a model directory under the user's home and may install Python packages, which are typical for a local model skill.

Assessment

This skill appears to do what it says (local ChatTTS). Before installing or running: 1) Verify and trust the mirror (https://hf-mirror.com) used to download model files—prefer official huggingface hosts if possible. 2) Inspect the ChatTTS package (and its source) before allowing automatic pip installs; consider installing dependencies in a virtualenv. 3) Be aware the scripts will download ~500MB to ~/.openclaw/ChatTTS/models and may overwrite/force-download files. 4) If you prefer safer operation, run download scripts manually and review their output, or change HF_ENDPOINT to the official huggingface.co endpoint. 5) If you are uncomfortable with network downloads and auto-install, do not run the scripts or run them in an isolated environment.

ChatTTS 本地语音合成

快速开始

使用方法

在消息中要求生成语音，例如：

• "用语音回复我"
• "把这段话读出来：[文字内容]"
• "生成语音：今天天气真好"

输出格式

使用 <qqvoice> 标签发送语音文件：

<qqvoice>/path/to/output.wav</qqvoice>

技术细节

模型信息

• 模型: ChatTTS
• 语言: 中文（主要优化）
• 采样率: 24000Hz
• 格式: WAV/MP3
• 运行: 本地 GPU/CPU

参数配置

参数	默认值	说明
speed	1.0	语速 (0.5-2.0)
pitch	1.0	音调 (0.5-2.0)
temperature	0.3	随机性 (0-1)
top_k	20	采样参数
top_p	0.7	采样参数

脚本调用

使用 scripts/tts.py 生成语音：

python3 scripts/tts.py "要转换的文字" --output output.wav

参数

• text: 要转换的文字（必填）
• --output: 输出文件路径
• --speed: 语速调节
• --pitch: 音调调节
• --seed: 随机种子（固定音色）

依赖安装

首次使用需要安装依赖：

pip install ChatTTS torch torchaudio

注意事项

1. 首次运行会下载模型（约 500MB）
2. GPU 加速可选，CPU 也能运行但较慢
3. 长文本建议分段处理（<500 字/段）
4. 音色由 seed 参数控制，固定 seed 可复现相同音色