RabiSpeech Qwen3-TTS
SkillMediaLets your agent generate spoken audio files in different languages, personas, or cloned voices using a local text-to-speech model.
Available today. Use it from your connected AI after setup.
No other account needed.
Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.
Then ask your AI: use the RabiSpeech Qwen3-TTS skill
About this skill
Generates persona-based, multilingual, or voice-cloned speech via RabiSpeech using local Qwen3-TTS 0.6B or 1.7B. Use when the user specifies Qwen3-TTS, Japanese/multilingual, local reference audio, a persona name, or model size comparison. Do not start the in-skill worker, OumuQ, or any Qwen/DashSco
What this skill tells your AI
The instructions your AI receives, as published by vb2250158/rabiroute in skills/qwen-tts-audio/SKILL.md and read by ahel’s review.
可用模型:
local-tts/qwen3-tts-0.6b-base:显存和冷启动较低,适合常规多语言对话。local-tts/qwen3-tts-1.7b-base:更大的本地模型,质量优先时选择。
流程
先读 TTS 路由。speechBaseUrl 必须从当前服务配置取得;下例中的 $speechBaseUrl 由该发现步骤赋值。
- 查询
GET <speechBaseUrl>/v1/models,确认模型已安装。 - 把 Rabi 人格目录名作为
voice;每次请求显式发送,不能继承上一会话角色。 - 使用
language指定目标语音语言,使用instructions提供简短风格/情绪说明。 - 对话用
play=true和session_id;测试用play=false保存 WAV。
$body = @{
model = 'local-tts/qwen3-tts-0.6b-base'
input = 'こんばんは。これはローカル音声テストです。'
voice = '<RoleId>'
response_format = 'wav'
language = 'ja'
instructions = '自然で落ち着いた会話調。'
play = $false
} | ConvertTo-Json -Compress
Invoke-WebRequest -Method Post -Uri ($speechBaseUrl.TrimEnd('/') + '/v1/audio/speech') -ContentType 'application/json' -Body $body -OutFile '.\qwen3-tts.wav'
模型加载、参考音频选择、缓存和 worker 生命周期全部由 RabiSpeech 管理。
Signals
- GitHub stars
- 502
- Last commit
- Sep 2026
Advanced
- Item type
- skill
- Key
qwen-tts-audio- Source
- github.com/vb2250158/rabiroute