hi. If i want to train text to Chinese speech audiolm , what confuse me is that pretrained hubert model is English-style, Does it affect my Chinese version training? Or i have to re-train hubert with my own large chinese dataset ? Appreciate !!!! @lucidrains

lucidrains / audiolm-pytorch

Implementation of AudioLM, a SOTA Language Modeling Approach to Audio Generation out of Google Research, in Pytorch

MIT License

2.33k stars 249 forks source link

hi. If i want to train text to Chinese speech audiolm , what confuse me is that pretrained hubert model is English-style, Does it affect my Chinese version training? Or i have to re-train hubert with my own large chinese dataset ? Appreciate !!!! @lucidrains #219

Closed hyhzl closed 11 months ago

glide-the commented 11 months ago

This should be the most accurate answer you can get

https://github.com/lucidrains/audiolm-pytorch/issues/80#issuecomment-1432293079

hyhzl commented 11 months ago

This should be the most accurate answer you can get

#80 (comment)

what? i had read hubert 、wav2vec 2.0 etc.

hyhzl commented 11 months ago

i found chinese hubert model,trained by tencent, but k-means is not provided, which it means it is not compatible with project ,that i could not get n_clusters and cluster_centers, what should i do ?

lucidrains commented 11 months ago

@hyhzl can you bring this to the discussions?