Getting started

Requirements

VoicePad currently requires Linux x86_64, a supported NVIDIA driver, and an NVIDIA CUDA GPU. The maintained hardware baseline is an RTX 3050 Laptop GPU with 4 GB physical VRAM. CPU-only ASR, AMD GPUs, and macOS are not supported.

Install from source

Package publication is currently frozen while the NVIDIA backend is qualified. For now, run VoicePad from a reviewed checkout:

mise install
uv sync --upgrade

No separate CUDA toolkit is required; the locked PyTorch dependencies include the reviewed CUDA runtime libraries. A persistent local uv tool installation has also been qualified, but is not yet the public distribution path.

Prepare the deployment

uv run voicepad prepare

The first preparation downloads and verifies approximately 2.5 GB of official Parakeet files and the official Silero wheel. Later startup is offline-capable.

Use the TUI

uv run voicepad

Wait for the NVIDIA device to show as ready. Press Space to begin recording and Space again to stop. VoicePad finalizes the WAV, drains bounded transcription work, writes Markdown, and copies only a complete result.

Use the CLI

uv run voicepad record start --duration 10
uv run voicepad transcribe path/to/existing.wav

Existing input audio is opened read-only. Output collisions fail rather than overwrite prior data.

Dictate from another application

Keep the TUI running, then bind voicepad toggle to a compositor-managed shortcut. Press it once to record and again to stop. See the global desktop shortcut guide for COSMIC and other Wayland desktops.