OpenAI STT

OpenAI's transcription model provides solid multilingual speech-to-text with optional real-time streaming mode.

Requires: OpenAI API key in Settings → Credentials

Models

Configuration

Code
{
  "transcriber": {
    "provider": "openai",
    "model": "gpt-4o-mini-transcribe",
    "language": "en",
    "detect_language": false,
    "prompt": null,
    "noise_reduction_type": "auto",
    "use_realtime": false
  }
}

Fields

Realtime Mode

When use_realtime: true, the OpenAI Realtime API is used for streaming transcription. Additional turn detection parameters become available:

Code
{
  "transcriber": {
    "provider": "openai",
    "model": "gpt-4o-mini-transcribe",
    "use_realtime": true,
    "turn_detection_type": "server_vad",
    "turn_detection_threshold": 0.5,
    "turn_detection_prefix_padding_ms": 600,
    "turn_detection_silence_duration_ms": 350
  }
}

Related

• Deepgram — better accuracy for most use cases
• Phantom STT — coming soon