Pretrained model

[yonzhao]

this is a good job!
1 . when are you going to release the pretrained model?
2. Could you provide the operation doc with us ? i can follow your step to use the wenet( data/train/test)

[wsstriving]

We will upload the pretrained model trained on voxceleb1, but we have to admit that this data itself is not big enough. So it would be better if you can train on larger data. Will be done in one or two days.

For the document, currently, you can refer to the readme file, which is quite detailed: https://github.com/wenet-e2e/wesep/tree/master/examples/librimix/tse/v2

[yonzhao]

Thanks for your quick reply! I have another question to ask you.
If i want to make the tse model better in mandarin, Do you have any Chinese data sets to recommend or other ways? As far as i know，most of reserchers train the TSE model in english data set.

[yezhangyinge]

We will upload the pretrained model trained on voxceleb1, but we have to admit that this data itself is not big enough. So it would be better if you can train on larger data. Will be done in one or two days.

For the document, currently, you can refer to the readme file, which is quite detailed: https://github.com/wenet-e2e/wesep/tree/master/examples/librimix/tse/v2

Hello, this is really a great work and useful to me. And I want to know when will you upload the pretrained model?

[SheenChi]

Great work, I also want to use the pretrained model to do some experiment. Can you provide the pretrained model? Thank you much @wsstriving

[mrjunjieli]

Pretrained models: https://modelscope.cn/datasets/wenet/wesep_pretrained_models/files

[wendongj]

Pretrained models: https://modelscope.cn/datasets/wenet/wesep_pretrained_models/files

thanks, hope see the code in this paper: "Multi-Level Speaker Representation for Target
Speaker Extraction" ^_^

[mrjunjieli]

We will release the code once it is accepted by ICASSP. Thanks for supporting our work.

[wendongj]

We will release the code once it is accepted by ICASSP. Thanks for supporting our work.

understand, you are welcome. waiting for you

[sangeet2020]

We will upload the pretrained model trained on voxceleb1, but we have to admit that this data itself is not big enough. So it would be better if you can train on larger data. Will be done in one or two days.

For the document, currently, you can refer to the readme file, which is quite detailed: https://github.com/wenet-e2e/wesep/tree/master/examples/librimix/tse/v2

Hi @wsstriving,
What is the total duration of the training data of this pre-trained model? Also, which dataset and what duration would you recommend to train one for superior performance?
Thank you