Are the pre-trained models available somewhere?

[ghost]

Many thanks!

[robin1001]

It's on developing now, will provide it once it is ready.

[whiteshirt0429]

we have uploaded some pre-trained models, the details can be seen in s0/README

[ghost]

Thanks,
Here https://github.com/mobvoi/wenet/blob/main/examples/librispeech/s0/README.md?
I don't see anything there.

[mikelei]

There are pre-trained models for AIShell in:
https://github.com/mobvoi/wenet/tree/main/examples/aishell/s0/README.md

Librispeech models need to be uploaded. @whiteshirt0429

[whiteshirt0429]

I have uploaded a librispeech u2 model，the model link is "http://mobvoi-speech-public.ufile.ucloud.cn/public/wenet/librispeech/20210122_unified_conformer_exp.tar.gz" and you can use git hash "1ea7515e4722f4e05a5985f587120c994a93694a" to load it.

[mikelei]

It seems oss is not accessible from outside China. Maybe put in some other places, such as dropbox?

[jprobichaud]

Thanks a lot for providing this! I see that in the pretrained model archives, the bpe models aren't available. If one wants to try these models on different audio files, we would need this file. Otherwise, we have to download all the training data and rebuild to bpe model.

[arkadyark]

Hey @jprobichaud - I was stuck on the same thing. After I looked a bit more, I was able to find a file (for instance in 20210215_unified_conformer_exp.tar.gz) called "train_960_unigram5000.model" which is the BPE model. This should work for you!