Skip to content

Voice Bar

The Voice Bar is a floating macOS widget that provides visual feedback and quick controls for VoiceLayer. Built with SwiftUI, it connects to the MCP server via Unix socket IPC.

Requirements

  • macOS 14+ (Sonoma or later)
  • Swift toolchain (Xcode or swift CLI)
  • VoiceLayer MCP server running

Quick Start

# Build and launch
voicelayer bar

# Stop
voicelayer bar-stop

The Voice Bar appears as a floating pill at the top of your screen.

Features

State Display

The pill changes appearance based on VoiceLayer state:

State Appearance
Idle Collapses to a small dot after 5s of inactivity
Speaking Expanded pill with teleprompter showing spoken text
Recording Pulsing waveform visualization driven by audio levels
Transcribing Loading indicator while whisper.cpp processes
Error Red indicator with error message

Teleprompter

During voice_speak, the pill expands to show the text being spoken:

  • Short text (8 words or fewer) renders centered
  • Long text uses a scrolling view with automatic scroll-to-current
  • Wider reading area (220pt, 11pt font) for comfortable reading
  • Punctuation-aware timing — pauses at periods and commas

Audio Waveform

During recording, a real-time waveform visualization shows microphone levels:

  • RMS audio levels received every ~100ms via socket events
  • Bar heights animated with spring physics
  • Visual confirmation that the mic is picking up audio

Click-to-Record

Tap the pill to start recording. When recording stops, the transcribed text is automatically pasted into the active application via Cmd+V (CGEvent).

Draggable Positioning

  • Drag the pill anywhere on screen
  • Position saved as screen percentage (survives resolution changes)
  • Restored on launch and when switching monitors

Idle Collapse

After 5 seconds of idle state, the pill collapses to a minimal dot. It expands automatically when any voice activity starts, or on hover.

Architecture

VoiceLayer MCP Server
    │
    │  /tmp/voicelayer.sock (NDJSON)
    │
    ▼
Voice Bar (SwiftUI)
    ├── NWConnection (Unix socket client)
    ├── @Observable VoiceLayerState
    ├── PillView (main UI)
    ├── TeleprompterView (text display)
    └── WaveformView (audio levels)

Socket Protocol

The Voice Bar communicates via NDJSON events over /tmp/voicelayer.sock:

Events (server → client):

Event Fields Description
state state, text? State change (idle, speaking, recording, transcribing, error)
speech text, index Text being spoken
audio_level level RMS 0.0-1.0, every ~100ms during recording
transcription text Transcription result
error message Error message

Commands (client → server):

Command Description
stop Stop current playback or recording
cancel Cancel current operation
replay Replay last TTS output
toggle Toggle TTS/mic on/off
record Start recording

Building from Source

bash flow-bar/build-app.sh
# Builds, bundles, codesigns, installs to /Applications/VoiceBar.app

The Voice Bar uses NSPanel with .nonactivatingPanel style so it never steals focus from your editor.