Deep Learning Homework – Korszerű Gépészek

Homework repository of team "Korszeru Gepeszek" for deep learning course: BMEVITMAV45 (2019/20/1) The project is carried out in Google Colaboratory using Python 3.7. Deep_Learning_Documentation.pdf contains the full documentation of the project.

Talker separation problem

The aim of our homework is to create a network that is capable of separating the speech of two speakers who speak simultanously. The method of recording is presumed to be single channeled (mono/monaural).

Data sets

The data source is the audiobook form of the Charles Dickens novel Hard Times, read by ten different speakers. To create input data we split and mix these signals on top of each other.

Dependencies

• tensorflow for the model, using tensorflow 1.x
• librosa, soundfile for audio processing
• h5py for saving the generated dataset in HDF5 format
• mir_eval for SDR calculation

Results

Output samples can be found in the model_example_outputs folder.