Show HN: Local audio transcription and speaker ID for Apple Silicon

A Python project that combines MLX Whisper for fast local transcription with pyannote.audio for speaker diarization, optimized for Apple Silicon.

• Fast local transcription using MLX Whisper (Apple Silicon optimized)
• Speaker diarization with pyannote.audio to identify different speakers
• Multiple output formats: TXT, SRT subtitles, JSON
• Privacy-focused: All processing done locally (only model downloads require internet)
• Robust error handling with fallback methods

• Python 3.8+
• Apple Silicon Mac (for MLX optimization)
• HuggingFace account and token with gated repository access

1. Install dependencies:

2. Get HuggingFace token:

   • Visit https://huggingface.co/settings/tokens
   • Create token with "Read access to gated repositories" permission
   • Accept user conditions for:
     • https://hf.co/pyannote/speaker-diarization-3.1
     • https://hf.co/pyannote/segmentation-3.0

3. Create .env file:

• TXT: Clean text with speaker labels
• SRT: Subtitle file with timestamps and speaker identification
• JSON: Full structured data with segments, timestamps, and metadata

• speech-to-text-fixed.py - Main script with speaker diarization
• transcribe_only.py - Simple transcription without speaker identification
• debug_pyannote.py - Debugging tool for pyannote issues
• speech-to-text.py - Original script (may have tensor size issues)
• test-mlx.py - MLX Whisper testing script
• CLAUDE.md - Development guidance for Claude Code

Common issues and solutions are documented in CLAUDE.md. For debugging pyannote issues, use: