Doubts related to bunched lpcnet paper and its implementation
Hi, I have been trying to evaluate the bunched lpcnet implementation with sample bunching and bit bunchinghttps://arxiv.org/pdf/2008.04574v1.pdf. Initially tried to modify the existing codebase for sample bunching but some layer related information is not clear in the paper as well. So is there someone working on this already? or atleast idea of the paper wch we can discuss?
Any leads are appreciated. Thank you Sagar
Did you read the papers about subband and multiband LPCNet ?
Are you referring to Featherwave https://arxiv.org/pdf/2005.05551.pdf ? Multiband linear prediction is used in this and I am currently going through the paper. Let me know if you are referring to something else.
are you working in the area of TTS ?
Yeah, Have started looking at new deep learning based methods
Hi, I have been trying to evaluate the bunched lpcnet implementation with sample bunching and bit bunchinghttps://arxiv.org/pdf/2008.04574v1.pdf. Initially tried to modify the existing codebase for sample bunching but some layer related information is not clear in the paper as well. So is there someone working on this already? or atleast idea of the paper wch we can discuss?
Any leads are appreciated. Thank you Sagar
I have implemented it with the 8bit ulaw and two samples one decode step. And synthesis speed can be 1.5X faster than the official. But sometimes, it will generate the click sounds.
A pytorch implement of Bunched LPCNet https://github.com/BridgetteSong/BunchedLPCnet