voice-bird-cli 0.3.3

Voice Bird CLI - local-first voice transcription TUI

Voice Bird CLI

Voice Bird CLI is a terminal voice transcription app. It runs locally by default with Whisper models, so local recordings stay on your machine. When you want hosted transcription, you can opt in per source to VoiceBird Web cloud mode.

Basic Flow

1. Pick a microphone, system-output loopback device, or app audio source.
2. Choose local or cloud mode for that source.
3. Press Enter to start a transcription slot.
4. Watch committed and tentative transcript text stream into the TUI.
5. In local mode, review session files under ~/voice-bird/sessions/<timestamp>-<source>/.

Local sessions contain:

File	Content
audio.wav	16 kHz mono recording
transcript.jsonl	Append-only transcript log, useful after crashes
transcript.json	Finalized transcript segments and metadata
transcript.txt	Plain-text transcript
meta.json	Device, source, model, engine, and duration

Getting Started

Install one of the CLI packages, then run:

voice-bird-cli

On first launch, Voice Bird picks a local Whisper model and downloads it into your OS cache directory. The default model is distil-small.en. Settings are stored in ~/.config/voice-bird/config.toml on Linux/macOS and %APPDATA%\voice-bird\config.toml on Windows.

Press m to change models. The picker includes nemotron-3.5-asr-streaming-0.6b, NVIDIA's latest Nemotron 3.5 ASR streaming model via the local parakeet-rs engine. Select it, let the package download/unpack, then start recording; the engine label and meta.json should show nemotron.

macOS users may need to grant Screen Recording permission for system or app audio capture. Apple Silicon users can optionally build the WhisperKit sidecar for ANE-accelerated local inference:

cargo run -p xtask -- build-sidecar

Without the sidecar, Voice Bird falls back to whisper-rs with whisper.cpp.

Install

Cargo installs the native Rust binary directly:

cargo install voice-bird-cli

PyPI installs a small wrapper that installs/runs the Cargo binary:

pipx install voice-bird-cli
# or
pip install voice-bird-cli

npm installs a small wrapper that installs/runs the Cargo binary:

npm install -g voice-bird-cli

From source:

git clone https://github.com/voice-bird/voice-bird-cli.git
cd voice-bird-cli
cargo install --path .

The npm and PyPI packages require Rust Cargo on the machine. Use the Cargo or source install when you want the simplest path.

Local And Cloud Modes

Free local mode is the default. It uses local Whisper inference through whisper-rs, or WhisperKit on macOS when the sidecar is available. Local mode does not send audio to a server and writes session artifacts to disk. Current local models are English-focused in the app flow.

Cloud mode streams audio to VoiceBird Web at wss://voicebird.app/api/audio/stream. It is opt-in, requires a Voice Bird API key, and is useful when local hardware cannot keep up or when you want cloud language support. Cloud recordings live in your VoiceBird Web account instead of the local sessions folder.

Your Voice Bird API key is stored in plaintext in config.toml; on Unix the app sets the file to 0600 best-effort.

Usage

voice-bird-cli                         # start the TUI
voice-bird-cli --recover <session-dir> # rebuild transcript.{json,txt} after a crash

Keys

Key	Action
↑ / ↓	Select device or app
← / →	Move between panes
Enter	Start the selected source
Space	Clear selected app pairing
Tab	Move between transcript slots
r	Refresh devices and apps
c	Toggle cloud mode for the focused source
l	Change language for cloud mode
m	Change model
e	Export the latest local transcript
p	Change local session path
x	Clear stopped transcript slot
q	Quit
?	Help

License

MIT