BUTSpeechFIT/VBx

Error: Empty input file in predict.py

Themba4Sho opened this issue · 3 comments

Themba4Sho commented

Hi there, I am trying to run the example script, I changed to the 8 kHz model and added a file in a new 8kHz folder where I put one example file. When I run the script, I get this error: Empty input file: "example/vad/9BHU18367S.lab".

I don't get this error when I run the 16 kHz model with the given example file. How do I generate the lab file? My assumption is that if I want to diarise, I just give the input file and model. Thanks.

fnlandini commented

Hi, the code in VBx assumes that you have already some voice activity detection (VAD) labels (contents of the .lab files) that mark the segments of speech (everything else is silence). For research purposes, sometimes we have oracle annotations about that (for example, like in the results listed in the README). For a real application, you would need to run VAD on your data. We have a very simple energy-based VAD in another branch that you could try to run this. However, I recommend you to use something better like the one from pyannote. I hope this helps.

Themba4Sho commented

Hi there, thank you for the response, I will do so.

fnlandini commented

I will close the issue now. If you find any trouble, feel free to reopen.