[hts-users:02585] Question about the inverse of msd stream variance
Hi, all
In HTS_engine, there is a step calculating inverse variance before the parameter generation. However, the inverse of msd stream variance is different from that of non-msd stream. There is a decision statement like: if the current frame is not the bound of msd frames or the variance is static, then the inverse is 1/vari. Otherwise, the inverse will be set zero.
What is interesting is that, as far as I checked, the dynamic variance of the msd bound frames happened to be rather small, which means that the inverse would be very large if not set to be zero.
Does this have a theoretical explanation about this phenomenon? Thanks very much!
Na Xingyu
您想拥有和网易免费邮箱一样强大的软件吗？