search

JasonSWFu / Quality-Net

Quality-Net: An End-to-End Non-intrusive Speech Quality Assessment Model based on BLSTM. (Interspeech, 2018, with Travel Grants)
88 stars 16 forks source link

Understanding how to extract PESQ #6

Open DavidHarar opened 3 years ago

DavidHarar commented 3 years ago

Hi, thanks for your code and article. I am trying to run Quality_net_training_Non_intrusive.py but I don't understand where I can get the pesq scores from. Can I use the pypesq package or there is another way to calculate them? I did not understand how read_pesq from MetricGAN works, nor how the files should look like. Can you please elaborate, or upload the data you used?

Thank you very much, David

DavidHarar commented 3 years ago

In addition, where is the corruption of files made? In the paper is written:

The speech utterances in the noisy set were obtained by corrupting the original speech utterances with 90 noise types, at eight SNR levels (from -10 dB to 25 dB with steps of 5 dB).

Can you please elaborate a little about this part?

Thanks again!