Voce
FRONT BURNER
Started June 2026
A macOS menubar app that learns your voice and filters incoming audio in real-time so only your speech comes through on calls.
From 0 to 1
Problem
Open-plan offices, shared homes and public spaces leak background voices into your calls. Existing noise cancellation removes noise, not other people speaking, and cloud-based voice filters add latency and privacy concerns.
Build
A macOS menubar app that learns your voice in a short enrollment step, then runs on-device speaker identification and real-time filtering through a BlackHole virtual audio device. Built in Rust with ONNX Runtime and CoreML acceleration on Apple Silicon, cpal for audio I/O and wry/winit for the UI.
Outcome
Real-time, fully on-device voice filtering: only your speech reaches the call, with zero latency and zero data leaving your machine.
Stack
Rust · ONNX Runtime · CoreML · cpal · BlackHole
Voce is a privacy-first, on-device macOS app that performs real-time speaker identification and voice filtering. It learns your voice during a short enrollment step, then uses that model to pass only your speech and silence everyone else’s — no cloud, no latency, no data leaving your machine.
Features
- On-device speaker enrollment and identification
- Real-time audio filtering via virtual audio device (BlackHole)
- CoreML acceleration on Apple Silicon
- Runs as a lightweight macOS menubar app
Tech Stack
Built in Rust with ONNX Runtime and CoreML for inference, cpal for audio I/O, and wry/winit for the menubar UI. Targets Apple Silicon (M1/M2/M3).
Links