Upload a clean 5–30 second clip of one speaker. Add a reference transcript (the exact words in the clip) and a language for a higher-fidelity clone — both optional but recommended. Vietnamese (vi) routes to the dedicated Vietnamese model. You can also do this in the UI: Voices → Register Voice.

curl -X POST https://siren.nopslabs.com/api/voices/register \
  -H "X-API-Key: YOUR_KEY" \
  -F "name=Triet — Studio" \
  -F "language=en" \
  -F "ref_text=The exact words spoken in the clip." \
  -F "file=@/path/to/reference.wav"
# → { "voice_id": "bacc...", "name": "Triet — Studio", "created_at": "..." }

A voice can store several style takes — e.g. one calm, one excited, one serious. Each is its own clip. At synthesis time you pick a styleId and the delivery inherits that take's emotion and phrasing. This is the most powerful lever — the quality of your reference takes is what gives genuine emotional range. Manage these in the UI under Voices → Manage styles.

curl -X POST https://siren.nopslabs.com/api/voices/VOICE_ID/styles \
  -H "X-API-Key: YOUR_KEY" \
  -F "label=Excited" \
  -F "ref_text=Words spoken in the excited take." \
  -F "file=@/path/to/excited_take.wav"
# → { "id": "style_2dcb...", "label": "Excited", ... }

# list / delete
curl https://siren.nopslabs.com/api/voices/VOICE_ID/styles -H "X-API-Key: YOUR_KEY"
curl -X DELETE https://siren.nopslabs.com/api/voices/VOICE_ID/styles/style_2dcb... -H "X-API-Key: YOUR_KEY"

Combine everything in one call to /api/audio/synthesize. All intonation fields are optional and stack: pick a styleId for emotion, add voice-design tags, and use markup in the script. (For timed subtitles, send srt instead of script — each cue is duration-locked to its window.)

curl -X POST https://siren.nopslabs.com/api/audio/synthesize \
  -H "X-API-Key: YOUR_KEY" -H "Content-Type: application/json" \
  -d '{
    "voiceId": "VOICE_ID",
    "styleId": "style_2dcb...",          // optional — clone from this expressive take
    "tts_language": "en",
    "tts_quality": "quality",
    "pitch": "high pitch",                // voice-design (validated, optional)
    "accent": "british accent",
    "whisper": false,
    "speed": 1.0,
    "script": "Wait [pause] this is the **only** way... really."
  }'
# → { "audioUrl": "https://...", "srtUrl": "https://..." }

Voice-design tags

Coarse, global shaping that layers on top of a clone. Values are validated — an unknown value returns 422 with the allowed list.

Prosody markup

Write these directly in your script for sentence-level phrasing. They compile to punctuation the model turns into real pauses — and work together with a cloned voice and a style.

Example: "Wait [pause] this is the **only** way... really." → spoken with a beat after "Wait", emphasis on "only", and a trailing hesitation before "really".