Data generation code

[pratheeksk]

hi,
I wanted to try out the code, how can I generate the data, is the code for data generation also uploaded to GitHub

[jnoelvictorino]

Hi,

I am also interested on how the mixing / data generation is done.

[divyeshrajpura4114]

Authors have adopted method from Single-Channel Multi-Speaker Separation using Deep Clustering to generate mixture of data. You can get script over here Deep Clustering. I hope this would help you.

[xin-h963]

@divyeshrajpura4114
Thank you for the link!
I can generate mixture of data but how can I get input feature, wiener-filter like mask, ideal binary mask, weight threshold matrix? (which is needed in data_utils.py)
I used TIMIT dataset.

[divyeshrajpura4114]

@xin-h963 T-F mask is just ratio of spectrograms of different speakers present in mixture. Please read the literature about Time-Frequency Mask. Below are some suggetion,

1. Time-Frequency Masking for Speech Separation and Its Potential for Hearing Aid Design.
2. On the Ideal Ratio Mask as the Goal of Computational Auditory Scene Analysis.