Transcription — Options

createTranscription(options) wires a provider to a recorder and exposes a single place to configure transports.

Top‑level options

type TransportKind = 'websocket' | 'http' | 'auto';

interface CreateTranscriptionOptions {
  provider: TranscriptionProvider;
  transport?: TransportKind; // default: 'auto'
  recorder?: Recorder; // pass to reuse an external recorder; otherwise the hook may create one

  // If using an internal recorder (in hooks):
  stages?: Stage[];
  preconnectBufferMs?: number; // buffer before connect, ms
  flushOnSegmentEnd?: boolean; // for HTTP: flush when VAD ends a segment

  connection?: {
    ws?: {
      silencePolicy?: 'keep' | 'drop' | 'mute';
      retry?: {
        enabled?: boolean;
        maxAttempts?: number;
        baseDelayMs?: number;
        factor?: number; // exponential backoff
        maxDelayMs?: number;
        jitterRatio?: number; // 0..1
      };
    };
    http?: {
      chunking?: {
        intervalMs?: number; // 0 = only on segment end
        minDurationMs?: number; // do not flush chunks shorter than this
        overlapMs?: number; // pad end with overlap
        maxInFlight?: number; // limit concurrent requests
        timeoutMs?: number; // per‑request timeout
      };
    };
  };
}

Notes

transport: 'auto' picks WS when available; otherwise falls back to HTTP.
For HTTP, typical “segment‑only” setup is flushOnSegmentEnd: true and intervalMs: 0.
For WS, use silencePolicy: 'drop' to cut bandwidth during silence; mute preserves cadence with zeroed frames.

Provider capabilities

Providers may implement only one method (WS or HTTP). The controller validates at runtime and throws if a method is missing.