VoicePad

VoicePad is a private, local-first dictation application for Linux systems with an NVIDIA GPU. It records durable WAV audio and transcribes it with NVIDIA’s official Parakeet TDT 0.6B v3 model through PyTorch FP16 CUDA.

Current production target

• Linux x86_64
• NVIDIA CUDA GPU in the tested 4 GB physical class or larger
• Official nvidia/parakeet-tdt-0.6b-v3 safetensors
• Official Silero VAD 6.2.1 on CPU
• Python 3.13 and uv

VoicePad provides no silent CPU ASR fallback. Unsupported hardware fails before recording begins.

Commands

voicepad prepare
voicepad transcribe recording.wav
voicepad record start
voicepad

Running voicepad without a command opens the TUI. Press Space to record or stop. The model remains resident between recordings.

Audio is persisted independently of inference. Existing WAV and Markdown files are never overwritten automatically.