Migrate from Resemble AI - Bland

Resemble sends a complete synthesis request with a request_id, then returns audio until audio_end. Bland uses a context_id for a turn and can accept that turn token by token. Start with the Realtime TTS Quickstart.

Message mapping

Resemble WebSocket	Bland `/v2/tts/ws`
`voice_uuid`	`init.voice`
`data` containing complete text or SSML	One or more `speak.text` deltas
`request_id`	`context_id`
`binary_response: true`	Binary audio is always used
`no_audio_header: true`	Realtime audio is always raw, without a WAV header
`audio_end`	`utterance_end`

for await (const token of llmTextStream) {
  ws.send(
    JSON.stringify({
      type: "speak",
      context_id: turnId,
      text: token,
    }),
  );
}

ws.send(JSON.stringify({ type: "end_of_turn", context_id: turnId }));

Compatibility notes

A Resemble project_uuid has no Bland equivalent.
Bland realtime supports PCM16 and 8 kHz mu-law. It does not support PCM24, PCM32, or MP3 on this WebSocket.
Bland does not currently return grapheme, phoneme, or word timestamps.
Resemble SSML and prompt markup do not map directly. Send plain text and use Bland voice controls where appropriate.
A new Bland context preempts the active turn. It does not run multiple request IDs concurrently.

For complete text and a WAV response, use Synthesize Speech (HTTP). For conversational input, keep one WebSocket open and use a new context for every LLM turn. See the Resemble streaming WebSocket reference and Bland turn concepts.

Docs for agents: llms.txt

⌘I