speaker cluster method

[Liujingxiu23]

"spectral clustering to the speech segments" is there any github code, link or paper？

[Liujingxiu23]

1.Did you use https://github.com/wq2012/SpectralCluster? In the README, it claims several methods, like base, auto-tune, which detail method did you use? 2.Could I use the same method to process gigaspeech? Do you have any exprience to process this dataset?

[tomasJwYU]

Hi, We use https://github.com/wenet-e2e/wespeaker/tree/master/examples/voxconverse/v1 for spectral clustering. More technical details can be found at https://arxiv.org/pdf/2309.13905.pdf. I will provide our speaker cluster results for gigaspeech in two days. (*￣︶￣)

[Liujingxiu23]

Hi, We use https://github.com/wenet-e2e/wespeaker/tree/master/examples/voxconverse/v1 for spectral clustering. More technical details can be found at https://arxiv.org/pdf/2309.13905.pdf. I will provide our speaker cluster results for gigaspeech in two days. (*￣︶￣)

Thank you for your reply! It's great to share gigaspeech result! I find your paper, "AUTOPREP", it's a great work!

For noise reduction, there is no office public tools for “Band-SplitRNN”? Can you recommend any other public tools？ Is https://github.com/facebookresearch/denoiser a good choice?

[tomasJwYU]

Thanks for the warm comment to AutoPrep~(^▽^)

For BSRNN you can refer to this repo https://gitlab.aicrowd.com/Tomasyu/sdx-2023-music-demixing-track-starter-kit But you need to set the hyper-parameter of the model to meet your own requirements

For gigaspeech speaker cluster results， please see https://tomasjwyu.github.io/AutoPrepDemo/

[Liujingxiu23]

@tomasJwYU thanks a lot!

[Liujingxiu23]

@tomasJwYU For gigaspeech， what is the cos_sim stands for？ the cos_sim of current wav segment to ？ I do not understand why most of the values are 1.0?