Transcription Settings

Project transcription settings control how speech is captured in the browser and how medical terminology is recognized during realtime and async transcription jobs.

Configure these in Project Settings → Transcription.

Background noise reduction

This toggle controls the browser microphone constraint noiseSuppression. It reduces ambient noise before audio reaches Scribeberry.

• Applies to browser clients using the SDK or WebSocket realtime flow
• Does not change server-side audio processing
• Disable only if your integration already applies its own noise filtering

Echo cancellation

Controls the browser echoCancellation constraint. Leave enabled for most clinical workstations and telehealth setups.

Recognition defaults

Client code can still pass explicit language or diarization options per session.

Medical dictionary

Improve recognition of specialty vocabulary by adding:

• Specialty — e.g. Dentistry, Family Medicine
• Clinic name — organization context for proper nouns
• Custom terms — one term per line (procedure names, product names, abbreviations)
• Additional context — short free-text guidance for uncommon phrasing

Dictionary content is applied server-side on each transcription session. It is not exposed to browser clients as raw configuration.

Size limit

The combined dictionary payload is limited to 10,000 characters. The console shows a live character count while editing.

SDK integration

When you mint a realtime token from your server, the response includes sessionDefaults:

{
  "token": "sb_rt_...",
  "expiresAt": "2026-08-06T20:00:00.000Z",
  "wsUrl": "wss://api.scribeberry.com/ws/realtime",
  "sessionDefaults": {
    "audioProcessing": {
      "noiseSuppression": true,
      "echoCancellation": true
    },
    "recognition": {
      "language": "en-US",
      "enableDiarization": true
    }
  }
}

The React hook useTranscription reads these defaults automatically before calling getUserMedia.

For custom integrations, fetch the token first, apply sessionDefaults.audioProcessing to microphone constraints, then connect to the realtime WebSocket.