Snowvoice Tts

本地中文语音合成工具，基于 Qwen3-TTS。当用户需要将文字转换为语音、克隆声音、 设计新音色或生成对话配音时触发。支持自然语言直接调用 snowvoice CLI 生成音频。 Keywords: TTS, 语音合成, 文字转语音, 声音克隆, 音色设计, 配音, snowvoice, 语音生成.

Security Scan

The skill's code, instructions, and requirements are internally consistent with a local SnowVoice-based TTS installer/CLI wrapper; the main risk is running the upstream repo's install script and downloading large models, which is expected for this purpose.

Purpose & Capability

Name/description (local SnowVoice TTS, voice cloning/design) match the provided code and runtime instructions: the skill discovers/installs a SnowVoice repo, manages model downloads, and wraps the snowvoice CLI.

Instruction Scope

SKILL.md and scripts instruct the agent to clone the upstream GitHub repo, run its install.sh, download ~3GB of models, and run the project's CLI. The instructions stay within TTS/installation scope but include executing install.sh from the cloned repo (see install_mechanism).

Install Mechanism

No packaged install spec, but init.py will git clone https://github.com/webkubor/snowvoice-studio and execute its install.sh (bash). Cloning from GitHub is expected, but executing a remote install script on the user's machine is a meaningful risk that users should review; model downloads (via modelscope CLI) will contact external model registries.

Credentials

The skill requests no environment variables, credentials, or unusual config paths beyond writing to user-home under ~/.snowvoice-studio (and an optional Desktop path). The filesystem access is proportionate for a local TTS installer.

Persistence & Privilege

always is false and the skill does not attempt to modify other skills or global agent settings. It will install files under the user's home directory (default ~/.snowvoice-studio) which is normal for this functionality.

Assessment

This skill is coherent for local TTS: it clones an upstream GitHub project and executes that project's install.sh, then downloads ~3GB of models and runs the installed CLI. Before installing, review the upstream repository (especially install.sh) to ensure you trust it; confirm you have sufficient disk space and bandwidth; consider running the setup manually in a sandbox or virtual environment rather than allowing an agent to run it autonomously; note SKILL.md claims macOS Apple Silicon but the scripts do not strictly enforce OS checks (they only check Python version and git). If you are uncomfortable with remote install scripts executing on your machine, do not run the automated setup and instead perform a manual, inspected installation.

本地中文语音合成 Skill，基于 SnowVoice Studio (Qwen3-TTS)。

触发条件

• "把这段话转成语音" / "用小烛的声音说你好"
• "克隆这个声音" / "设计一个温柔的女声"
• "生成一段配音" / "帮我朗读这段文字"

环境初始化

首次使用前，必须确保 SnowVoice Studio 已安装。 Agent 应自动检测并引导安装。

检查环境

python3 scripts/init.py status

一键安装（推荐）

python3 scripts/init.py setup

这会自动完成：git clone → venv 创建 → 依赖安装 → 模型下载（约 3GB）

分步安装

python3 scripts/init.py install          # 仅安装软件
python3 scripts/init.py download-model Base-1.7B          # 下载克隆模型
python3 scripts/init.py download-model VoiceDesign-1.7B   # 下载设计模型

安装路径

默认安装到 ~/.snowvoice-studio。如果开发者路径 ~/Desktop/personal/tts 存在也会被识别。

执行流程

1. 检查环境：确认 snowvoice 可用（自动发现路径）
2. 识别意图：判断克隆/设计/列表
3. 提取参数：文字内容 + 音色（自然语言解析）
4. 构建并执行命令：python -m cli.app clone/design <args>（在 snowvoice 项目目录下）
5. 返回结果：音频文件完整路径

核心命令

克隆（最常用）

# 在 snowvoice 项目目录下执行
python -m cli.app clone <persona_key> "要合成的文字"
python -m cli.app clone <persona_key> "文字" --tone "温柔" --emotion "Happy"

音色设计

python -m cli.app design <voice_name> "建模短句" --tone "描述"

音色列表

python -m cli.app voice list

音色速查

简称	persona_key	说明
顾栖月	gu_qiyue	默认音色
小烛	candy	小烛原版
小烛傲娇	candy_cool	傲娇大小姐
小烛腹黑	candy_mischievous	腹黑小恶魔
王爷沉稳	王爷-儒武沉稳	儒武风格
王爷冷峻	王爷-冷峻锋压	冷峻风格
星栀	星栀-暧昧撩人	AI女友
夜棠	夜棠-午夜耳语	AI女友
朝朝	朝朝-元气阳光	AI女友
宁观尘	ning_guanchen	男声
江湖老人	jianghu_laoren	老年男声
搞笑男	zhou_xingchi	港式无厘头

完整列表运行 snowvoice voice list 查看。自然语言映射见 scripts/tts_skill.py 的 PERSONA_MAP。

注意事项

1. 首次使用必须初始化：python3 scripts/init.py setup
2. 模型下载约 3GB：首次可能需要 10-30 分钟
3. 仅支持 macOS Apple Silicon（MPS 加速）
4. 执行超时：合成一条语音约 10-30 秒，已设 5 分钟超时
5. 输出路径：默认在 snowvoice 项目的 out/ 目录