Speaker Diarization - Audio Length Limit

[sjsakshi]

Hi,

Thanks for the prompt reply on helping me out with diarization with unknown speakers.

I just had a question - is there a limit on the length of audio file that can be processed for speaker diarization , or any speaker model in general with nemo.

If I have files with 1.5 hrs length, can they be effectively diarized?

Thanks!!

[tango4j]

Speaker diarization can be done on an audio file of 90mins, but it could take few minutes to get the result since clustering algorithm is relying on eigenvalue decomposition (O(n^3) complexity)

accuracy wise, diarization accuracy is not necessarily dependent on the length of audio.

Divide and conquer feature for faster clustering is WIP.

[sjsakshi]

Thanks so much for the reply..

While I was trying further, I got another similar question..

If a file is of very short duration 2-4 seconds, will the embeddings be extracted accurately.??

Is there a minimum limit of audio required for effective speaker embedding extraction?

Thanks!!

[nithinraok]

Technically you can extract speaker embedding with any length of audio. I would recommend a minimum speech length of 2 sec to get better representations of the speaker. For diarization purposes, you may use smaller window lengths as well if you use it with multi-scale embeddings extraction.

[github-actions[bot]]

This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.

[github-actions[bot]]

This issue was closed because it has been inactive for 7 days since being marked as stale.