Generate conversation with timestamps

Synthesizes a multi-turn conversation and returns a JSON envelope with base64 audio and word-level timestamps mapped back to each originating turn.

Authorization

bearerAuth

AuthorizationBearer <token>

API key

In: header

Request Body

application/json

TypeScript Definitions

Use the request body type in TypeScript.

mode?string

Discriminator hinting this is a conversation request; optional since the endpoint already implies it.

Value in"conversation"

model?string

Provider/model used by every flat turn that doesn't supply its own model. Set this OR per-turn model — not both.

Match^[a-z0-9-]+\/[a-zA-Z0-9._-]+$

gapMs?integer

Silence inserted between consecutive turns, in milliseconds. Defaults to 0 (no gap).

Range0 <= value

volumeDbfs?number

Target peak loudness in dBFS (negative; e.g. -16). Each turn is normalized to this level before stitching.

Rangevalue <= 0

providerOptions?

Free-form passthrough applied to every turn's upstream provider call. Per-turn providerOptions override these.

output?||

Output container format. Accepts either a string shorthand ("wav" | "mp3" | "pcm") or an object ({ format, bitrate? }); bitrate is only valid with mp3 and defaults to 96 kbps when omitted. Defaults to the provider's native stitched format (typically wav). Note: pcm is headerless raw audio — consumers must know the sample rate out-of-band. Response Content-Type reflects the chosen format.

timestamps?string

Whether to return word-level timestamps. 'on' returns alignment (default), 'off' skips alignment entirely.

Value in"on" | "off"

pronunciations?

Pronunciation overrides applied to every turn's text before synthesis: saved dictionary IDs and/or inline rules.

moderation_ruleset_id?string

UUID of the org moderation ruleset to evaluate this request against. When omitted, the org's default ruleset applies.

Match^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i

turns*array<||>

Ordered list of speaker turns to synthesize and stitch into one audio file. Must contain at least one turn.

Items1 <= items

Response Body

`application/json`

`application/problem+json`

`application/json`

curl -X POST "https://example.com/v1/audio/conversation/with-timestamps" \  -H "Authorization: Bearer $SPEECHBASE_API_KEY" \  -H "Content-Type: application/json" \  -d '{  "turns": [    {      "mode": "inline",      "model": "string",      "voice": "string",      "text": "string"    }  ]}'

{
  "audio": "string",
  "mediaType": "string",
  "warnings": [
    "string"
  ],
  "timestamps": [
    {
      "text": "string",
      "start": 0,
      "end": 0,
      "turnIndex": 0
    }
  ]
}

{
  "type": "string",
  "title": "string",
  "status": 0,
  "detail": "string"
}

{
  "type": "string",
  "title": "string",
  "status": 0,
  "detail": "string"
}

{
  "error": {
    "code": "content_moderation_blocked",
    "message": "string",
    "reason": {
      "type": "error_fail_closed"
    }
  }
}

Generate conversation with timestamps

Authorization

Request Body

Response Body

200application/json

400application/problem+json

401application/problem+json

422application/json

`application/json`

`application/problem+json`

`application/problem+json`

`application/json`