Broadcast TTS

How to generate

• POST /v1/audio/speech with model=clavue-tts
• input = the script to read (plain text, not lyrics tags)
• voice = Uncle_Fu (default, steady male) or Vivian (crisp female)
• instructions = locked news-desk diction (send it; the API fills the default if omitted)
• Timeout 120s is enough. Do not send audio_duration, seed, or reference_audio

curl -sS https://api.clavue.com/v1/audio/speech \
  -H "Authorization: Bearer $CLAVUE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model":"clavue-tts",
    "input":"晚间天气。城区有短时阵雨，气温二十二到二十六度。请携带雨具。",
    "voice":"Uncle_Fu",
    "instructions":"请用新闻播音腔，吐字清楚，语速中等，数字读完整，专名不吞字。不要口语，不要气声，不要唱歌。",
    "response_format":"wav"
  }' --output news.wav --max-time 120

Recommended settings

Lock the voice and the diction. Do not scan voices. Casual Bot UI speech can stay on the internal short TTS path — this product is for news desk and scripted reads.

• voice: Uncle_Fu (default) · Vivian · Serena
• instructions: news-desk lock (see curl). Do not write song captions here
• response_format: wav · read the real WAV duration

{
  "model": "clavue-tts",
  "voice": "Uncle_Fu",
  "response_format": "wav"
}

Mix with music

Dialogue is clavue-tts. Beds and opening stings are clavue-music with Instrumental, no vocals. Two calls, then mix in your product. Do not ask the song model to sing the news.

Limits

One infer per replica. Timeout 120s. Clone fields are rejected. This is not a singing model and not a replacement for clavue-music.