VoxForge
Hi,
In HTK, HCompV can be used to estimate the global mean and covariance of a set of training data. But the defaut configuration of HCompV can give only the global mean and covariance matrix for a single Gaussian. I would like to estimate the global means and covariance matrices when I have a mixture of Gaussian. Do anyone want knows how to do this with HCompV?
(I have change the proto to a mixture of Gaussian but the HCompV gave the same means and covariance matrices for all the mixture. I think this is not correct. Normally, the mean and covariance matrix in each Gaussian of the mixture are different.)
Thanks for your help.
Best regards,
Thanh
> I have change the proto to a mixture of Gaussian but the HCompV gave the same means and covariance matrices for all the mixture. I think this is not correct.
I think it is correct.
> Normally, the mean and covariance matrix in each Gaussian of the mixture are different.)
Since there is no additional information provided, there is no way to distinguish mixtures. So no reason for them to be different on the initial stage.
Later after HInit they will be different of course