Question about the 'real' subfolder

[ahikaml]

Hi, Thank you for this dataset. I have two questions about the real sub-folder in this dataset.

1. Are the sweep files played by the device speakers or by some external speaker?
2. Is there any place in which I find the clean near-end signals of the '*nearend_singletalk_mic.wav' files, that can be used later as a target?

Thanks

[andosa]

1. The sweep files are played by whatever the user is using for audio playback, in general it is the device speakers.
2. Not sure what you mean by clean signals. All near-end files are real recordings of a user reading a prompt

[ahikaml]

Thanks for the quick response.

Regarding (2) - I meant to the near-end signal without background noise and room+device responses, similar to the files in the folder ..\synthetic\nearend_speech or the *_nearend_singletalk_lpb files in the test folders. But if all near-end files in the 'real' folder were recorded by a real person (and not by a speaker simulating a person), then it's quite understood why this folder doesn't contain such files.. .

As far as I understand models that combine echo cancellation and noise reduction/dereverberation jointly will not be able to be trained using this folder 'as is' because many files there contain background noise and reverb. Please correct me if I'm wrong
Thanks