live-translator

Records computer or microphone audio in real time and translates it to Brazilian Portuguese using OpenAI Whisper and GPT-4o mini.

How it works

1. Audio is captured from the selected input device in 30-second chunks
2. Silent chunks are skipped automatically
3. Each chunk is transcribed by Whisper (language-agnostic)
4. The transcription is translated to pt-BR by GPT-4o mini

Requirements

• Python 3.13+
• uv
• An OpenAI API key

Installation

git clone <repo>
cd live-translator
uv sync

Create a .env file in the project root:

OPENAI_API_KEY=sk-...

Capturing computer audio (macOS)

To translate audio playing on your computer (not just your microphone), you need a loopback device.

1. Install BlackHole 2ch (free)
2. Open Audio MIDI Setup (Spotlight → "Audio MIDI Setup")
3. Click + → Create Multi-Output Device
4. Check both BlackHole 2ch and your speakers/headphones
5. In System Preferences → Sound → Output, select the new Multi-Output Device

Audio will now play through your speakers and be routed into BlackHole simultaneously.

Note: volume keys stop working with a Multi-Output Device. To fix it, check "Use this device for sound output" on the speakers entry inside the Multi-Output Device in Audio MIDI Setup.

Usage

# List available input devices
uv run live-translator --list-devices

# Translate from BlackHole (computer audio)
uv run live-translator --device-name blackhole

# Or select by device index
uv run live-translator --device 3

# Adjust chunk size (default: 30s)
uv run live-translator --device-name blackhole --chunk 10

Options

Flag	Short	Default	Description
--list-devices	-l	—	List available input devices and exit
--device	-d	system default	Input device index
--device-name	-n	—	Select device by name substring (e.g. blackhole)
--chunk	-c	30	Audio buffer length in seconds
--silence-threshold	-s	0.005	RMS level below which chunks are skipped