preprocessing
:
DataEncoder
: Responsible for creating and saving the pre-encoded versions of the audio files used for training.EncoderDataset
: Responsible for loading the raw audio files, noise files, applying preprocessing steps (such as cutting to equal length, applying bandpass filter) and creating the noisy audio files.
train
:
EncodedDataset
: Responsible for loading the pre-encoded input and target files for training the model.LatentNetwork
: Contains the structure of the latent network used in the audio denoising problem.LatentTrainer
: Responsible for training the latent network with the pre-encoded dataset.utils
:ModelSaveHandler
: Utility class for saving model checkpoints.
normal_6sec.mp4
- Noise: Hitting table
original1.mp4
cleaned1.mp4
- Noise: Pressing plastic bag
original2.mp4
cleaned2.mp4
- Noise: Hitting glass with a fork
original3.mp4
cleaned3.mp4
- Noise: Playing pop music
original4.mp4
cleaned4.mp4
- Noise: Playing classical music
original5.mp4
cleaned5.mp4
- Noise: Playing dubstep music