AudioFlex: Python Library for Audio Stretching

Introduction

Welcome to AudioFlex, a comprehensive Python library dedicated to audio stretching and manipulation. This project stems from a gap I found in the realm of audio processing tools – the lack of a pure Python library for intricate audio stretching algorithms. AudioFlex is developed to demystify the complexities often shrouded by extensive C++ implementations, making these algorithms accessible and understandable to a broader audience.

Example

import soundfile as sf  # Example dependency for reading and writing audio
from audioflex.wsola import WSOLA

chunk_size = 1024
frame_size = 512
sound_path = 'path/to/input/sound.wav'
output_path = 'path/to/output/file.wav'

with sf.SoundFile(sound_path) as sound, \
        sf.SoundFile(output_path, 'w', samplerate=sound.samplerate, channels=sound.channels) as output:
    wsola = WSOLA(channels=sound.channels, chunk_size=chunk_size, frame_size=frame_size)
    while True:
        frame = sound.read(frames=chunk_size, dtype='float32').T
        processed_chunk = wsola.process(frame, stretch_factor=0.7)
        output.write(processed_chunk.T)

Motivation

The inspiration for AudioFlex originated from the need for a Python-centric approach to audio stretching. Existing solutions heavily rely on C++ code, creating a barrier for those seeking to understand the underlying mechanics. This project aims to simplify this process, offering a suite of algorithms in Python that are both educational and functional. In the future I want to create some music software / instruments using these algorithms.

Features

• Pure Python Implementation: Unlike other libraries that depend on C++ bindings, AudioFlex is written entirely in Python, ensuring ease of understanding and modification.
• Comprehensive Algorithm Suite: The library includes various algorithms for audio stretching, pitch shifting, and time-scaling, providing users with multiple tools for audio manipulation.
• Educational Resource: The algorithms are presented in a way that emphasizes learning and understanding, making this library an excellent resource for students and enthusiasts alike.

Reference Material

This project heavily references the work and teachings of AudioSmith, particularly the lecture "Four Ways To Write A Pitch-Shifter - Geraint Luff - ADC22." The lecture and accompanying C++ code have been instrumental in shaping the understanding and implementation of the algorithms in this library. For a deeper dive into the topic, we highly encourage watching the lecture here.

Requirements

Please ensure you have the necessary dependencies installed. A requirements.txt file is included for convenience. Install the dependencies using:

pip install -r requirements.txt

Dependencies:

• numpy (version ~1.26.3)
• scipy (version ~1.11.4)

Usage and Modules

AudioFlex is designed for ease of use with detailed documentation. Key modules include:

• wsola.py: Implements the WSOLA algorithm for time-scale modification.
• buffer_handlers.py: Manages audio data efficiently using numpy arrays.
• overlap_add.py: Incorporates the Overlap-Add method for audio stretching.
• And more in the makes...

Contribution

Contributions to AudioFlex are welcomed and appreciated. Whether it's improving the code, adding new features, or enhancing documentation, your input is appreciated.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

Special thanks to AudioSmith and Geraint Luff for their invaluable insights and contributions to the field of audio processing. Their work has been a guiding light in the development of this library.