README.md

SEGAN Datasets

A SeganAugDataset is constructed from some path to clean .wav files and a path to noise .wav files. While training, it will add noises from noisy audio files into clean audio signals on the fly according to your configuration 😙

A SeganDataset is constructed from a directory containing clean .wav and a directory containing noisy .wav.

Inputs

class SeganAugTrainDataset(BaseDataset):
    def __init__(self,
                 stage: str,
                 clean_dir: str,
                 noisy_dir: str,
                 speech_config: dict,
                 cache: bool = False,
                 shuffle: bool = False)
                 
class SeganTrainDataset(BaseDataset):
    def __init__(self,
                 stage: str,
                 clean_dir: str,
                 noisy_dir: str,
                 speech_config: dict,
                 cache: bool = False,
                 shuffle: bool = False):

Outputs when iterating for training

(clean_wav_slices, noisy_wav_slices)

Outputs when iterating for testing

(clean_wav_path, noisy_wavs)