Hi Sobh,The GV weight controls the trade-off between generating high-probability trajectories and generating trajectories with high-probability global variance values. See any paper on parameter generation considering global variance for the precise formula (for example, Toda 2007).
In practice the effect of GV parameter generation with the conventional weight (1.0 in HTS) is typically to set the global variance of all generated trajectories to be almost exactly equal to a fixed value. This is mentioned, for example, in "An introduction to speech synthesis" by Simon King. This means that if you vary the GV weight slightly, as you have done, you still get essentially the same trajectories, still with global variance almost exactly equal to that fixed value. If you choose a very large positive value (or possibly very small, if I've got it the wrong way round) then you should start to notice a difference -- the generated trajectories will start to look like the ones you get when GV is disabled.
Cheers, Matt On 16/05/13 08:13, Ibrahim Sobh wrote:
Dear I tried to change the gvWeight from 1.0, 0.7, 0.5 and 0.2 in the HTS demo. However there is no clear difference between generated wav files! Could you please give me the mathematical formula for the gvWeight = 1.0 in the HTS demo. I need to know the effect of changing gvWeight Best Regards Sobh