OpenAIWhisperEngine

OpenAI Whisper API transcription engine.

Note: This is NOT true real-time streaming. OpenAI's Whisper API processes complete audio files, so this engine buffers audio and sends it in chunks rather than doing true low-latency streaming.

This engine is best suited for:

• Post-recording transcription

• High accuracy requirements

• Short audio clips

Example

val engine = OpenAIWhisperEngine(apiKey = "your-openai-api-key")

// Best used with complete recordings
val recording = recorder.getRecording()
recording?.asAudioFlow()?.transcribe(engine)?.collect { result ->
    println("Transcription: ${result}")
}

Browsers (JS / WasmJS)

OpenAI's API does not send Access-Control-Allow-Origin, so direct calls from a browser are blocked by CORS. Stand up a thin backend that forwards the multipart upload to OpenAI and point endpointUrl at it. Your backend keeps the API key; pass any auth headers it requires via additionalHeaders (and leave apiKey blank if your backend doesn't proxy the bearer token).

val engine = OpenAIWhisperEngine(
    apiKey = "",
    endpointUrl = "https://my-app.example.com/api/transcribe",
    additionalHeaders = headersOf("X-App-Token", "client-secret"),
)

Parameters