MusicGen-Minimal v1.1.0

Text-to-music generation that actually works. No bullshit, just music.

What This Does

• 🎵 Generates music from text descriptions
• 🚀 Extended generation beyond 30 seconds (NEW)
• 📦 Batch processing from CSV files (NEW)
• 🎧 MP3 output support (NEW)
• 📊 Real-time progress bars (NEW)
• ✅ Uses Facebook's MusicGen model
• 💻 Works on CPU (slow) or GPU (fast)

What This Doesn't Do

• No web interface (yet)
• No real-time generation
• No style transfer
• No MIDI export
• No audio conditioning

Installation

⚠️ Reality Check: This package downloads 2.7GB of dependencies (PyTorch, etc.) on first install.

# Basic installation
pip install musicgen-minimal

# With MP3 support  
pip install 'musicgen-minimal[audio]'

First install includes:

• PyTorch (2.1GB) - ML framework
• Transformers (400MB) - HuggingFace models
• Other dependencies (300MB)
• Total: ~2.7GB download, ~7GB disk space

Optional MP3 support:

• Install pydub: pip install pydub
• Install ffmpeg: https://ffmpeg.org/download.html
• Without these, falls back to WAV output

First use downloads model:

• Small model: 300MB (recommended)
• Medium model: 1.5GB (better quality)
• Large model: 3.2GB (best quality)

Usage

Basic Generation

# Generate 10 seconds of music
musicgen generate "upbeat jazz piano" --duration 10

# Generate 2-minute track (uses extended generation)
musicgen generate "epic orchestral symphony" --duration 120

# Generate with MP3 output (requires pydub)
musicgen generate "lofi hip hop" --output beats.mp3

# Disable progress bars
musicgen generate "ambient nature sounds" --no-progress

Batch Processing

# Create sample CSV template
musicgen create-sample-csv

# Process multiple tracks from CSV
musicgen batch playlist.csv

# Parallel processing with 4 workers
musicgen batch jobs.csv --workers 4

Python API

from musicgen import quick_generate

# Generate 60-second MP3
quick_generate(
    "cinematic trailer music",
    "trailer.mp3", 
    duration=60,
    format="mp3",
    bitrate="320k"
)

Performance (Brutal Honesty)

CPU Performance (Most Users)

• Model loading: ~30 seconds
• Generation speed: 0.1x realtime (10x slower)
• 30-second song: ~5 minutes
• 2-minute song: ~20 minutes
• Memory usage: ~4GB RAM

GPU Performance (CUDA)

• Model loading: ~5 seconds
• Generation speed: 2-5x realtime
• 30-second song: ~10 seconds
• 2-minute song: ~40 seconds
• Memory usage: ~4GB VRAM

⚠️ WARNING: Long generations on CPU will test your patience!

Why So Simple?

Because it actually works. The original project had 343 files and didn't run. This has 4 files and ships music.

What's New in v1.1.0

Extended Generation

• Generate music longer than 30 seconds (up to unlimited duration)
• Automatic segment blending with crossfade
• Progress tracking for multi-segment generation

Batch Processing

• Process multiple tracks from CSV files
• Parallel generation with multiprocessing
• Mixed format support (WAV and MP3 in same batch)

MP3 Output

• 75-90% smaller files than WAV
• Configurable bitrate (128k-320k)
• Graceful fallback if dependencies missing

Progress Bars

• Real-time progress for all operations
• ETA calculation for long generations
• Optional --no-progress for automation

Contributing

Keep it simple. If your PR adds more than 50 lines, it's too complex.