Intervox

Intervox is a macOS desktop app that interprets live speech and exposes the translated voice as a virtual microphone for any apps.

Speak normally. Intervox listens through your real microphone, translates what you say, and makes the translated voice available as Interpreter Mic. Zoom, Google Meet, QuickTime, and any app that can choose a microphone can use it like a regular input device.

What It Does

• Interprets live speech and sends the translated voice through a virtual microphone.
• Lets meeting apps hear the translated voice instead of the original mic.
• Supports silence, pass-through, translated voice, and translated voice mixed with the original.
• Shows captions, latency, connection state, usage estimates, and driver status.
• Keeps the OpenAI API key and local app data on the Mac.

How It Works

Intervox runs as a Tauri desktop app with a small macOS audio driver. The app captures a selected microphone, sends speech to OpenAI Realtime with the user's own API key, receives translated audio, and writes it to the virtual microphone.

The product name shown in the UI is Interpreter Mic. macOS and some audio selectors may show the underlying device as Intervox.

Requirements

• macOS 14 Sonoma or later.
• Rust 1.94.0, pinned by rust-toolchain.toml.
• Xcode Command Line Tools.
• Node.js 24.2.0 and pnpm 10.21.0.
• cmake and ninja for the audio driver.
• OpenAI API key for live translation.
• Apple Developer ID Application certificate and notarytool credentials for signed release builds.

Quick Start

Install dependencies:

pnpm install

Run the fast checks:

cargo test --workspace
pnpm test
pnpm check

Run the app in development mode:

pnpm tauri dev

When setup is incomplete, Intervox opens the first-run onboarding flow.

Build

Build a local unsigned development app:

pnpm run build:app:dev

Build the signed and notarized release app:

pnpm run build:app

build:app is the release path. It requires Developer ID and notarytool credentials because the app bundle includes the audio driver.

Driver

Build the driver:

scripts/build_driver.sh

Sign and notarize it for release or full local acceptance:

scripts/sign_driver.sh
scripts/notarize_driver.sh

Install it locally:

INTERVOX_ASSUME_YES=1 sudo bash scripts/install_driver.sh

The app also supports privileged driver install, update, and uninstall from onboarding and the Status pane.

Local Data

Intervox stores settings and the OpenAI API key in:

~/Library/Application Support/app.intervox.desktop/config.json

The config directory uses user-only permissions, and the config file is written with mode 600. The key is never returned to the frontend by the config IPC command.

Other local files live in the same app-data directory:

File or directory	Purpose
usage.json	Local month and lifetime usage estimate.
transcripts/*.jsonl	Per-session transcript history when transcript saving is enabled.
connection.log	Capped connection lifecycle log for troubleshooting.

Docs

Document	Purpose
docs/ARCHITECTURE.md	Runtime model, component boundaries, privacy model, and audio-driver invariants.
docs/DEVELOPMENT.md	Local setup, driver lifecycle, build/test commands, and probes.
docs/STATUS.md	Current implementation status and acceptance checklist.
docs/RUNBOOK-acceptance.md	Manual A1-A12 acceptance flow for real hardware and meeting apps.