Does dataset in synthetic folder contain only double talk scenarios？

[cjf1699]

Hi,
I found that all files in datasets/synthetic/nearend_mic_signal subfolder are mixtures of near end speech and echo signal of far end speech, without any files which contain only echo signal (simulating far end single talk) or near end speech (simulating near end single talk) . But I do not understand why, is it enough to only consider double talk scenarios when training an AEC model?
Thanks.

[sphaye]

"But I do not understand why, is it enough to only consider double talk scenarios when training an AEC model?"
No, you should train an AEC model with an appropriate ratio of Nearend singletalk, farend singletalk and doubletalk scenarios.
For the data provided, you can always simulate the farend singletalk by just using the echo signals instead of nearend_mic_signal, and similarly for the nearend singletalk.

BTW: do you have access to the synthetic_fullband dataset? I am unable to access that link and I need it for my experiments. Would be grateful if you could provide a secondary link if you have it downloaded!