API 레퍼런스 · Media
Conversations
Several voices, one scene: each line spoken by its own voice, returned as one MP3.
The voice studio's Conversation mode: send the lines in order, each with its voice, and the response body is ONE MP3 (audio/mpeg) with the lines spoken one after the other. It lands in your Library as one voice line, the script as its prompt.
Up to 40 lines and 5,000 characters in all. Priced exactly as the lines would be one by one — per started block of 100 characters, per line — and charged once: if any line fails, the whole conversation is refunded. The X-Credits-Spent header says what it cost.
Voices are the ones Text-to-speech takes: roster ids, your cloned voices, public voices, or char:<id>. On eroq Voice One, [bracket] delivery tags work line by line.
요청 본문
The conversation in order, up to 40 lines: [{ "voice": "…", "input": "…" }] — each line's voice (as in Text-to-speech) and its text.
eroq-voice-one (default, most expressive) or eroq-voice-turbo (faster, cheaper).
Playback speed 0.7–1.3, 1 = natural, for every line.
0 (steady) → 1 (theatrical), for every line.
Library folder the conversation lands in — one of yours (GET /v1/folders). Omitted = the Library root.
A flow to run on the conversation once it lands — one that starts « After a render » of voice lines (GET /v1/flows). See Flows.
Private mode: returned and kept nowhere — no Library copy, and the ledger shows the script as stars.
curl https://eroq.ai/v1/audio/dialogue \
-H "Authorization: Bearer $EROQ_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "eroq-voice-one",
"lines": [
{
"voice": "aria",
"input": "You came back. [whispering] I told you not to."
},
{
"voice": "orion",
"input": "You also told me the door would be locked."
}
]
}' \
--output speech.mp3응답
(binary MP3 body — audio/mpeg)