菜单
Text-to-speech using Qwen3-TTS CustomVoice MLX model. Supports 9 speakers and multiple emotion/style instructions. Optimized for Apple Silicon. Use when user wants audio speech output.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use this skill whenever the user wants iOS-style scroll fades, gradient masks, scroll-edge overlays, or top/bottom indicators on a scrollable list, dropdown, sidebar, modal, chat log, or panel. It implements a reusable wrapper-based scroll fade mask pattern across React, Vue, Svelte, and vanilla JavaScript, and it is also the right skill when the user says the fade masks broke scrolling, do not show, or do not match the background.
Bootstrap, operate, and troubleshoot `nelvko/clash-for-linux-install` on Linux hosts. Use when Codex needs to initialize Clash or Mihomo on a machine that may not already be installed, follow a daily-use SOP with `clashctl` and `clashsub`, inspect or edit `.env`, `resources/mixin.yaml`, or `resources/runtime.yaml`, manage subscriptions, Tun mode, Web UI secrets, upgrades, and uninstall, or quickly diagnose startup, port, service, subscription-conversion, or network-access problems. This skill vendors a trimmed upstream project copy under `scripts/upstream-project/` and should first try fetching the latest GitHub repository, then fall back to copying the bundled project files when GitHub access is slow or unavailable.
Operate the bundled PV Tool kinetic typography web app for lyric videos, promotional videos, and motion-graphics text overlays. Use this whenever the user mentions PV Tool, wants to run or preview the app locally, build it for distribution, experiment with PV templates and effects, or work with lyric text, media backgrounds, audio-reactive overlays, or LRC input inside PV Tool.
Use this skill whenever the user mentions Atlassian Design System, Atlaskit, Jira-style UI, Confluence-style UI, ADS components, design tokens, xcss, @atlaskit/css, primitives, navigation system, or wants to build/refactor React interfaces to match Atlassian patterns. This skill helps choose the right ADS component, combine multiple ADS patterns, and implement code grounded in the mirrored documentation under references/docs.
Reuse local Youtu-Tip GUI capabilities through a safe adapter CLI so OpenClaw/Codex-style agents can inspect desktop GUI state and perform guarded single-step actions on macOS.
Operate and control Claude Code CLI from within Clawdbot. Allows spawning Claude Code processes, executing commands in Claude Code, managing sessions, and deploying projects using Claude Code's capabilities.
| name | qwen-tts |
| description | Text-to-speech using Qwen3-TTS CustomVoice MLX model. Supports 9 speakers and multiple emotion/style instructions. Optimized for Apple Silicon. Use when user wants audio speech output. |
Text-to-speech using Qwen3-TTS CustomVoice model, running locally on Apple Silicon via MLX.
mlx-community/Qwen3-TTS-12Hz-0.6B-CustomVoice-4bit (4-bit quantized, ~600MB)cd /path/to/skills/skills/qwen-tts
python3 -m venv venv
source venv/bin/activate
模型从 ModelScope 镜像下载（国内更快）：
pip install -r scripts/requirements.txt
首次运行时会自动下载模型。如需提前下载：
source venv/bin/activate
export HF_ENDPOINT="https://hf-mirror.com"
python3 -c "from mlx_audio.tts.utils import load_model; load_model('mlx-community/Qwen3-TTS-12Hz-0.6B-CustomVoice-4bit')"
source venv/bin/activate
python3 scripts/tts.py "你好，这是一段测试语音。" --output /tmp
# Should produce a WAV file in /tmp/
| 问题 | 解决方法 |
|---|---|
ModuleNotFoundError: mlx | 确认使用 Apple Silicon Mac，MLX 不支持 Intel Mac |
| 模型下载缓慢 | 设置 export HF_ENDPOINT="https://hf-mirror.com" |
| 内存不足 | 4-bit 模型约需 1.5GB 内存，关闭其他大型应用 |
| 无声音输出 | 检查输出文件是否为 0 字节，可能是文本过短 |
⚠️ 以下为示例默认值。请根据实际使用场景修改 speaker 和 instruct。
Serena（示例）撒娇语气（示例）Chinese1.0| Speaker | Language |
|---|---|
Serena | Chinese |
Vivian | Chinese |
Uncle_Fu | Chinese |
Eric | Chinese |
Dylan | Chinese |
Ryan | English |
Aiden | English |
Ono_Anna | Japanese |
Sohee | Korean |
撒娇语气 — coquettish冷静分析 — calm analysis惊讶 — surprised兴奋 — excited神秘 — mysterious开心 — happy委屈 — wronged/sadAlso supports free-form natural language instructions, e.g. 用特别愤怒的语气说.
# Default settings
python3 scripts/tts.py "你好！"
# Custom speaker
python3 scripts/tts.py "Hello!" --speaker Ryan --language English
# Custom emotion
python3 scripts/tts.py "其实我真的有发现..." --instruct 冷静分析
# Full customization
python3 scripts/tts.py "哥哥，你回来啦！" \
--speaker Serena \
--instruct 撒娇语气 \
--speed 1.0
# Custom output directory
python3 scripts/tts.py "测试" --output /tmp
# Skip auto-cleanup of old files
python3 scripts/tts.py "测试" --no-cleanup
~/tts-output/ (override with $QWEN_TTS_OUTPUT_DIR)tts_{timestamp}_{index}.wav--no-cleanup)