lm-audioslicer 0.1.8

Tool for slicing long audio files for datasets

AudioSlicer

A Python tool for automatically slicing audio files based on silence detection, designed for creating ML datasets and processing large audio collections.

Features

• Automatic silence detection using RMS analysis
• Batch processing of WAV and MP3 files
• Random word generation for unique file naming
• Configurable silence thresholds and segment lengths
• Maintains original sample rate

Installation

You can install audioslicer either via pip or by cloning the repository:

Using pip

pip install lm_audioslicer

From source

git clone https://github.com/LarsMonstad/audioslicer.git
cd audioslicer
pip install -r requirements.txt

PIP usage

lm_audioslicer /path/to/audio/folder [options]

Usage

python -m audioslicer.slicer /path/to/audio/folder [options]

Options

• -t, --threshold: Silence threshold in dB (default: -40.0)
• -l, --min_length: Minimum segment length in ms (default: 5000)
• -i, --min_interval: Minimum silence interval in ms (default: 300)
• -s, --hop_size: Analysis window size in ms (default: 10)
• -m, --max_silence: Maximum silence to keep in ms (default: 1000)

Example

python audioslicer.slicer -m ./my_audio -t -35 -l 3000

PIP example

lm_audioslicer ./my_audio -t -35 -l 3000

How It Works

1. Analyzes audio using RMS (root mean square) to detect silence
2. Identifies silence regions below the threshold
3. Splits audio at optimal points within silence regions
4. Generates unique filenames using random pronounceable words
5. Saves segments in a new 'sliced_audio' directory

Output

• Creates 'sliced_audio' folder in the input directory
• Naming format: originalname_randomword_number.wav
• Preserves original audio quality and sample rate

Technical Notes

• Minimum segment length prevents creation of tiny clips
• Hop size affects precision vs processing speed
• Maximum silence parameter trims long silence periods
• Uses librosa for robust audio processing

Acknowledgments

• Inspired by GUI Audio Slicer

License

MIT License

---

Made by Lars Monstad