Transcribe Audio Skill

Fast local audio transcription with speaker diarisation. Outputs a transcript with speaker labels, automatically identifying speaker names from context where possible.

Backends:

• Parakeet + FluidAudio (default): Fast, local, runs on Apple Silicon. Transcription + speaker identification.
• AssemblyAI (cloud): Use only when user explicitly requests "AssemblyAI", "cloud transcription", or for non-English audio.

Prerequisites

For local transcription (default)

• parakeet-mlx at ~/.local/bin/parakeet-mlx
• fluidaudio CLI at ~/.local/bin/fluidaudio
• ffmpeg for audio format conversion (if needed)

If Parakeet is not installed:

uv tool install parakeet-mlx