Rotor-informed Wave-U-Net for Speech Enhancement

PyTorch implementation of Wave-U-Net for speech enhancement with additional CSV file with rotor rps time series.

Dependencies

librosa, numpy, torch, scipy, tqdm, soundpy, json5, pesq, pystoi, torchmetrics, soundfile

Clone

git clone https://github.com/gullogullo/Rotors-informed-Wave-U-Net-for-Speech-Enhancement.git

Usage

There are the SEGAN model, the Wave-U-Net model, and the rotor-informed Wave-U-Net model

• Entry file for training Wave-U-Net models: train.py
• Entry file for enhance noisy speech with Wave-U-Net models: enhancement.py and enhancement_rotors.py
• Entry file for training SEGAN model: segan_main.py
• Entry file for enhance noisy speech with SEGAN model: segan_enhancement.py

Examples

MMGG0_SI2339_Noisy.mov.webm

MMGG0_SI2339_Enhanced.mov.webm

MTRR0_SA2_Noisy.mov.webm

MTRR0_SA2_Enhanced.mov.webm