Skip to main content
The output_format query parameter selects the encoding and, on sync and async text to speech, the output sample rate and bitrate:
An encoding on its own keeps Breeze’s native 24000 Hz output. Adding a sample rate in Hz resamples to that rate, and mp3, aac, and opus take a further bitrate in kbps. HTTP streaming accepts only the bare pcm, mp3, and wav encodings; when omitted, streaming defaults to pcm to reduce time to first audio. Realtime text to speech over WebSocket does not accept output_format. It always streams pcm_s16le, 24000 Hz, mono, 16-bit audio as binary frames.
Breeze generates native 24000 Hz audio. A profile that names a higher sample rate resamples that output for compatibility with downstream editors, encoders, and media pipelines. It does not add acoustic bandwidth or improve the source quality.

​
Supported encodings

​
Sample rate and bitrate profiles

Every output is mono. WAV and raw PCM are signed little-endian 16-bit. Sample rates are accepted at 8000, 11025, 12000, 16000, 22050, 24000, 32000, 44100, and 48000 Hz. Opus encodes only at 8000, 12000, 16000, 24000, and 48000 Hz. MP3 takes the MPEG layer III bitrates for the chosen sample rate: 32, 40, 48, 56, 64, 80, 96, 112, 128, 160, 192, 224, 256, and 320 kbps at 32000 Hz and above; 8, 16, 24, 32, 40, 48, 56, 64, 80, 96, 112, 128, 144, and 160 kbps below it. AAC takes 8 to 320 kbps and Opus 6 to 510 kbps. wav, pcm, and flac do not take a bitrate. A value Breeze cannot deliver exactly returns BAD_REQUEST naming the reason rather than falling back to a nearby format, so a request never returns audio at a rate or bitrate other than the one it asked for. The HTTP streaming endpoint rejects any value carrying a sample rate. For async delivery, GET /v1/generation-jobs/{generation_job_id} reports the requested value as output_format and the delivered encoding, sample rate, channel count, and bitrate as audio_format.

​
Choosing a format

  • End-user playback in a browser or app: prefer mp3.
  • One-shot low-latency playback: use the HTTP streaming endpoint with the default pcm response for lower TTFA.
  • Realtime conversation: use the WebSocket realtime endpoint. The audio format is fixed to pcm_s16le, 24000 Hz, mono, 16-bit.
  • File-oriented streaming workflows: request wav or mp3.
  • Archival or post-processing: use wav or flac for lossless non-streaming output.
  • A fixed rate or bitrate required downstream: name it, for example wav_48000, pcm_16000, or mp3_44100_192, with sync or async text to speech.

​
Use formats with

SDK quickstart

Generate and save your first MP3 with the Python or TypeScript SDK.

Text to speech

Pass encoding-only values on sync, async, and streaming requests, or name a sample rate and bitrate on sync and async requests.

Convert text to speech

See where output_format fits among the request parameters, with SDK examples.

Streaming

Use streaming-compatible formats for lower-latency playback.

Realtime text to speech

Build realtime conversation audio with fixed PCM binary frames.

CLI text to speech

Save generated audio from the command line while prototyping voices and formats.