fishaudio/audio-preprocess

What it solves

It provides a set of tools for preparing and cleaning audio data, which is essential for training high-quality audio AI models. It handles common preprocessing tasks like removing background noise (vocal separation), normalizing volume, and converting formats.

How it works

The project is a command-line tool (fap) that executes various audio processing scripts. It can convert video or audio files to wav format, separate vocals from other sounds, slice audio into smaller segments, match loudness across different files, and perform transcription using tools like FunASR.

Who it’s for

Developers and researchers working with audio datasets who need to automate the cleaning, slicing, and transcription of raw audio files before they are used in machine learning pipelines.

Highlights

• Vocal Separation: Separates speech from background noise or music.
• Automatic Slicing: Breaks long audio files into manageable pieces.
• Loudness Matching: Ensures consistent volume levels across a dataset.
• Transcription: Supports generating .lab files and integrating with FunASR for automated transcription.