LMX Voice & Audio

Opta LMX exposes OpenAI-compatible audio endpoints for speech-to-text (STT) and text-to-speech (TTS) on Apple Silicon.

Overview

Voice and audio capabilities are served under /v1/audio/* and can be used directly by clients that already speak the OpenAI API contract.

Compatibility

Endpoint shape is aligned with OpenAI-style audio routes so existing tooling can integrate with minimal change.

Speech to Text

Transcription is available via /v1/audio/transcriptions. Upload audio (for example .wav, .mp3, .m4a) as multipart form data.

Transcribe audio with Whisper on LMX

curl -X POST http://lmx-host.local:1234/v1/audio/transcriptions \
  -F "file=@sample.wav" \
  -F "model=mlx-community/whisper-base"

Text to Speech

Speech synthesis is available via /v1/audio/speech. LMX streams generated audio from the configured TTS model.

POST /v1/audio/speech

{
  "model": "mlx-audio/kokoro-82m",
  "input": "Opta runtime is healthy and ready.",
  "voice": "af_heart",
  "response_format": "wav"
}

Generate speech audio from text

curl -X POST http://lmx-host.local:1234/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"model":"mlx-audio/kokoro-82m","input":"Hello from Opta LMX","voice":"af_heart"}' \
  --output speech.wav

Runtime Notes

Audio endpoints run on the same LMX service instance as chat completions.
Model availability and latency depend on Apple Silicon memory headroom.
Use /healthz and /readyz before long-running audio jobs.

Monitoring

Overview

LMX Voice & Audio

Overview

Speech to Text

Text to Speech

Runtime Notes

On this page