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

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

MMGG0_SI2339_Noisy.mov.webm

MMGG0_SI2339_Enhanced.mov.webm

MTRR0_SA2_Noisy.mov.webm

MTRR0_SA2_Enhanced.mov.webm

Name		Name	Last commit message	Last commit date
Latest commit History 9 Commits
__pycache__		__pycache__
config		config
dataset		dataset
model		model
trainer		trainer
util		util
.DS_Store		.DS_Store
.gitignore		.gitignore
README.md		README.md
dataset_creation.py		dataset_creation.py
enhancement.py		enhancement.py
enhancement_rotors.py		enhancement_rotors.py
model_rotor_graph		model_rotor_graph
model_rotor_graph.png		model_rotor_graph.png
segan_data_preprocess.py		segan_data_preprocess.py
segan_enhancement.py		segan_enhancement.py
segan_main.py		segan_main.py
segan_model.py		segan_model.py
train.py		train.py