qalarc.com / projects / qalarc-voice

Qalarc Voice

A project by Qalarc — AI systems studio, Sydney, Australia. Built, shipped and maintained by Qalarc.

Voice layer for the terminal with speaker recognition, voice cloning, and wake word activation

A voice layer for the terminal: speaker recognition, voice commands and text-to-speech, wired directly into the shell.

Speak commands instead of typing them; the layer recognises who is speaking, executes terminal input, and reads output back with TTS. It is the studio's testbed for voice as a first-class terminal interface — the same ideas later grew into gmux's on-device Whisper voice control.

Built for hands-busy workflows: servers across the room, notes while reading, accessibility.

Capabilities

• Faster-Whisper STT with streaming

• VibeVoice TTS at ~300ms latency

• Speaker identification via Resemblyzer

• Voice cloning from short samples

• Wake word: 'Hey Qal', 'Hey Claude'

• TUI, Web, and Daemon modes

Tags

Status: in-progress · First built: 2025-01-01 · Last updated: 2025-04-01