Does silero-vad contain any noise filtering algorithm?

[balazsgonczy]

Hi,

I would like to apply your algo on a Speaker Identification task on noisy recordings. Have you tested it on noisy data? If not then what noise filtering algo do you recommend? I have already tried "nr.reduce_noise" but it slows down the process insanely.

Bw,

Balázs

[snakers4]

Hi,

We have designed VAD to work with reasonably noisy data. Typically it requires fiddling with hyper-params a bit like setting longer minimal speech duration and higher threshold.
As for noise filtering / audio enhancement - this is still work in progress for us.

[snakers4]

As usual, try plotting the chart with your data to see how well VAD performs. The VAD utils have a flag for that

[balazsgonczy]

Maybe can you recommend a library or a specific function? Sry I am totally new to audio files, I used to working on tabular data and images. :D Alexander Veysov ***@***.***> ezt írta (időpont: 2022. febr. 25., P 8:53):

…

Hi, We have designed VAD to work with reasonably noisy data. Typically it requires fiddling with hyper-params a bit like setting longer minimal speech duration and higher threshold. As for noise filtering / audio enhancement - this is still work in progress for us. — Reply to this email directly, view it on GitHub <#172 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMVH6HIENP75E3GHHGLD7ADU44YQFANCNFSM5PJRPLAQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[snakers4]

the only thing that comes to mind is a denoiser by facebook
but it has some limitations, like poor scripts in the repo, limited sampling rate, etc

[balazsgonczy]

Thx I think Ill only stick to ur algo, cuz its doing its job 99.75% of the time. Alexander Veysov ***@***.***> ezt írta (időpont: 2022. febr. 25., P 9:42):

…

the only thing that comes to mind is a denoiser by facebook but it has some limitations, like poor scripts in the repo, limited sampling rate, etc — Reply to this email directly, view it on GitHub <#172 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMVH6HMKKL2JNEDPJWF2VSLU446IFANCNFSM5PJRPLAQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you authored the thread.Message ID: ***@***.***>