search

rrbluke / CDEC

Cross-Domain Echo Controller
MIT License
32 stars 17 forks source link

Can the frequency-domain block-based AEC automatically handle single talk and double talk situations? #2

Open Captain2xxx-coder opened 2 years ago

Captain2xxx-coder commented 2 years ago

Hi, I think this is a remarkable job! I have a question about the frequency-domain block-based AEC. Is the AEC method you used able to handle the doubletalk case automatically? Or do you have to separate the far-end singletalk and doubletalk cases during pre-processing and manually stop the filter coefficient update during the doubletalk case?

Captain2xxx-coder commented 2 years ago

Hi, I think this is a remarkable job! I have a question about the frequency-domain block-based AEC. Is the AEC method you used able to handle the doubletalk case automatically? Or do you have to separate the far-end singletalk and doubletalk cases during pre-processing and manually stop the filter coefficient update during the doubletalk case?

From generate_cache.py， I think linear AEC can automatically deal with single and double talk situations. Thank you for your project, it’s very useful for me!

w17786138647 commented 2 years ago

Hi, I think this is a remarkable job! I have a question about the frequency-domain block-based AEC. Is the AEC method you used able to handle the doubletalk case automatically? Or do you have to separate the far-end singletalk and doubletalk cases during pre-processing and manually stop the filter coefficient update during the doubletalk case?

hello，I saw that the author used the train-hard folder under the AEC-Challenge dataset in the code, but I did not find this folder in this dataset, did you see it? could you please give me a reply,thanks very much!

rrbluke commented 2 years ago

The AEC does not need a VAD, as you can see from the implementation, and the paper being referenced there.

rrbluke commented 2 years ago

train_hard contains the hardest files from the dataset, which have been selected in terms of SNR.

w17786138647 commented 2 years ago

it's so honored to receive your reply. i will read the code more carefully, thanks , best wishes !

w17786138647 commented 2 years ago

train_hard contains the hardest files from the dataset, which have been selected in terms of SNR.

Excuse me, is this code complete？I didn't see any information about adding data to the train_hard folder (the train_hard folder is empty at first, that is, 0 pieces of data, which brought errors to the operation of the code). If is complete, could you please tell me where is it? I'm so sorry to disturb you. And hope that I can get your reply. Thanks very mush , Best wishes!

M-Z-Yi commented 1 year ago

Hi, I think this is a remarkable job! I have a question about the frequency-domain block-based AEC. Is the AEC method you used able to handle the doubletalk case automatically? Or do you have to separate the far-end singletalk and doubletalk cases during pre-processing and manually stop the filter coefficient update during the doubletalk case?

From generate_cache.py， I think linear AEC can automatically deal with single and double talk situations. Thank you for your project, it’s very useful for me!

Hello, I'm so sorry to disturb you,from the generate_cache.py,I don't understand how linear AEC automatically deal with single and double talk situations.,can you tell me something about it?I'm looking forward to your reply.