[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:02591] Re: How exactly do we get LSP?


After checking the Makefile in data directory, I realized that after MGCEP (with gamma=-1) there is a lpc2lsp execution which convert Wraped-LPC to LSP. Then I do the following test:
1. Use MGCEP with alpha=0.42 and gamma=-1 to get Wraped-LPC. Use LPC2LSP to get LSP. Convert LSP back to LPC. Use POLEDF to filter using the resulting LPC coefficients. The voice sounded really wierd.
2. Chang alpha to 0. Do the exact steps above and the resulting voice sounds correct.
Here is the question. Is there a Wraped-LPC filter, which could use the Wrap-LPC the do all pole filter?
Thanks!

At 2010-08-26 09:59:17,"那兴宇" <nxy-yzqs@xxxxxxx> wrote: >Hi , all >I am getting more confused these days about LSP. >First, I configure the HTSdemo by "MGCLSP=1, GAMMA=1" to get LSP coefficients. But the synthesis part in the Train.pl did get silence results. So I synthesize them independently after the training was done. The results sound normal but there are always noise at the beginning and ending of the setences. > >Then I turn to test LSP in SPTK. As I read the tutorial slides by Prof. Tokuda, it was said that the coefficient we get by setting gamma = -1 is LPC ( wraped with alpha !=0 and linear with alpha ==0). In my understanding, the coefficient I got was actually Wraped-LPC rather then LSP. But in the synthesis part, gamma was used to determine MGC or MGC-LSP. And I didnt find the proper setting of training part to get LSP, both in the slide and the INSTALL documents. > >My question is, what exactly do we get by setting GAMMA to 1 and MGCLSP to 1 both? Wraped-LPC or LSP? >



您想拥有和网易免费邮箱一样强大的软件吗?

Follow-Ups
[hts-users:02592] Re: How exactly do we get LSP?, Heiga ZEN (Byung Ha CHUN)
References
[hts-users:02590] How exactly do we get LSP?, 那兴宇