Voice Input

The Amurg UI supports voice dictation with two backends. The default mode uses the browser's built-in Web Speech API and requires no setup. For private, offline speech recognition, you can connect a local Whisper server.

Voice Modes

Switch modes via the gear icon next to the microphone button. Settings are saved in localStorage under the key amurg-voice.

How to Use

The microphone button supports two interaction styles:

Edit before send

Transcribed text is appended to the message input field — it is never sent automatically. You can review, edit, or add to it before pressing send. While you speak, a real-time interim preview is shown above the input field.

Visual Feedback

Browser Mode

The default mode uses the browser's SpeechRecognition API (or webkitSpeechRecognition on Safari). It requires no server or configuration.

Local Whisper Mode

For private, offline speech recognition, you can run a Whisper-compatible ASR server and point the UI at it. Audio never leaves your machine.

Setup

1. Run a Whisper ASR server that accepts WebSocket connections and receives audio/webm chunks.
2. Click the gear icon next to the microphone button in the Amurg UI.
3. Select Local Whisper.
4. Enter the WebSocket URL (e.g. ws://localhost:8000/asr).

Protocol

The UI streams audio to the Whisper server in 250ms chunks using MediaRecorder with audio/webm;codecs=opus format. The server is expected to respond with JSON messages containing transcription results.

Expected server responses

The UI looks for a text or transcript field in the JSON response. It also recognizes partial/interim results via buffer, segments, is_final, and type: "partial" fields.

// Partial transcription (shown as interim preview)
{"type": "partial", "text": "hello wor"}

// Final transcription (appended to input field)
{"text": "hello world", "is_final": true}

// Alternative field names also accepted
{"transcript": "hello world"}

Settings Storage

Voice settings are persisted in localStorage under the key amurg-voice as a JSON object:

{
  "mode": "browser",
  "whisperUrl": "ws://localhost:8000/asr"
}

Troubleshooting