Fast Audio Processor for LLM Transcription

Simple, fast audio processing that chunks audio files and applies optional speed adjustments for efficient LLM transcription.

Features

• ✅ Fast Processing - No complex smart detection, just efficient chunking
• ✅ Auto Queue Management - Processes most recent file in pending/ folder
• ✅ Multiple Audio Formats - Supports OGG, MP3, WAV, M4A, FLAC, AAC
• ✅ Customizable Chunks - Set chunk duration (default: 17 minutes)
• ✅ Speed Adjustment - Optional speed modification (default: 1.0x - no change for fastest processing)
• ✅ Auto File Management - Moves completed files to completed/ folder
• ✅ Optimized Output - 64kbps MP3, 16kHz mono for transcription efficiency

Directory Structure

audio-process/
├── fast_processor.py      # Main processing script
├── requirements.txt       # Python dependencies
├── README.md             # This file
└── files/
    ├── output/           # Processed chunks go here
    └── queue/
        ├── pending/      # Drop audio files here
        └── completed/    # Processed files moved here

Installation

Prerequisites

Install FFmpeg (required by pydub):

macOS:

brew install ffmpeg

Ubuntu/Debian:

sudo apt update && sudo apt install ffmpeg

Python Dependencies

pip install -r requirements.txt

Usage

Quick Start

1. Drop your audio file in files/queue/pending/
2. Run the processor:

   python3 fast_processor.py

3. Find your chunks in files/output/[filename]/

Command Options

# Default: 17-minute chunks, no speed change (fastest)
python3 fast_processor.py

# Custom chunk duration (10 minutes)
python3 fast_processor.py -d 10

# Custom speed (1.5x faster)
python3 fast_processor.py -s 1.5

# No speed adjustment
python3 fast_processor.py -s 1.0

# Combine options
python3 fast_processor.py -d 15 -s 1.25

Command Line Options

• -d, --duration: Chunk duration in minutes (default: 17)
• -s, --speed: Speed multiplier (default: 1.05)
• --base-path: Custom path to audio-process directory

Example Output

For a 1-hour recording named meeting.ogg:

🚀 FAST Audio Processor
📁 Pending: /path/to/pending
📁 Output: /path/to/output
⚡ Mode: FAST (no smart chunking)
🔍 Processing: meeting.ogg

🚀 FAST Processing: meeting.ogg
📊 Chunk duration: 17 minutes
⚡ Speed factor: 1.05x
📥 Loading audio file...
✅ Loaded successfully
⏱️  Duration: 60.0 minutes
🔪 Creating 4 fixed chunks

⚙️  Processing chunk 1/4...
   🏃‍♂️ Applying 1.05x speed (pitch preserved)...
   ✅ meeting_chunk_01.mp3 (16.2min, 2.3MB)

⚙️  Processing chunk 2/4...
   🏃‍♂️ Applying 1.05x speed (pitch preserved)...
   ✅ meeting_chunk_02.mp3 (16.2min, 2.3MB)

⚙️  Processing chunk 3/4...
   🏃‍♂️ Applying 1.05x speed (pitch preserved)...
   ✅ meeting_chunk_03.mp3 (16.2min, 2.3MB)

⚙️  Processing chunk 4/4...
   🏃‍♂️ Applying 1.05x speed (pitch preserved)...
   ✅ meeting_chunk_04.mp3 (9.5min, 1.4MB)

🎯 Complete! 4 chunks in /path/to/output/meeting
📦 Moved to completed: meeting.ogg

🎉 SUCCESS!
   📁 Output: /path/to/output/meeting
   📊 Chunks: 4
   📦 Completed: /path/to/completed/meeting.ogg

Output Files

Files are saved with the naming pattern:

• filename_chunk_01.mp3
• filename_chunk_02.mp3
• filename_chunk_03.mp3
• etc.

Each chunk is:

• Sped up to your specified factor (default 1.05x) with pitch preserved
• Optimized for transcription: 64kbps, 16kHz, mono
• Small file size for efficient API uploads
• Properly named with zero-padded numbers for correct ordering

Workflow

1. Record your audio (any format: OGG, MP3, WAV, M4A, etc.)
2. Drop in pending folder: files/queue/pending/
3. Run processor: python3 fast_processor.py
4. Upload chunks to your transcription service
5. Original file automatically moves to completed/

Performance

• Fast Processing: No complex analysis, just efficient chunking
• Large Files: 1-hour audio processes in ~1-2 minutes
• Small Output: Optimized for transcription APIs (file size limits)
• Reliable: Simple, robust processing with good error handling

Transcription Tips

• 17-minute chunks work well with most LLM transcription services
• 1.05x speed saves time while maintaining natural pitch and accuracy
• Use 1.0x speed for difficult audio (accents, technical terms)
• Smaller chunks (10-15 min) for better error recovery
• Files are optimized for OpenAI Whisper, Assembly AI, etc.

Troubleshooting

File won't load?

• Check if FFmpeg is installed: ffmpeg -version
• File might be corrupted - try a different file

Processing too slow?

• Use smaller chunks: -d 10
• Skip speed adjustment: -s 1.0

Need different output quality?

• Edit the export parameters in fast_processor.py