Re: [Kaldi-developers] help regarding Kaldi hmm topology

[Daniel P. <dp...@gm...>]

If you look in the standard recipes, there is a prototype topology
somewhere e.g. in conf/topo.proto or somewhere like that that you could
edit if you want to have more than 3 states.
Regarding things like adding states for silence and short pause-- it sounds
to me like you are trying to "literally translate" a HTK recipe.  This is
not a good idea, as Kaldi is a different toolkit with different ways of
doing things.  It will just do the right thing with regard to the number of
Gaussians per state, so you don't have to really do anything.
Dan


On Thu, May 2, 2013 at 8:58 AM, Arif Khan <ife...@gm...> wrote:

> Hi,
>
> I want to run Aurora 2.0 experiments with Kaldi.  For that I need to model
> 18 state hmm for digits. and 5 state hmm for silence and 3 state hmm for
> short pause (between two consecutive digits).
>
> My question is:
> a) How we tie/share a state among different hmm models.
>
> b) I want to use iterative training, therefore will have to manipulate hmm
> in the middle of training for adding states for short pause and silence and
> increasing Gaussian mixtures per state, so How
> we modify hmm during the middle of training. In HTK we use HHED module to
> do this.
>
> Any help or reference material would be highly appreciated.
>
>
> Best regards,
> Arif
>
>
>
> ------------------------------------------------------------------------------
> Get 100% visibility into Java/.NET code with AppDynamics Lite
> It's a free troubleshooting tool designed for production
> Get down to code-level detail for bottlenecks, with <2% overhead.
> Download for free and get started troubleshooting in minutes.
> http://p.sf.net/sfu/appdyn_d2d_ap2
> _______________________________________________
> Kaldi-developers mailing list
> Kal...@li...
> https://lists.sourceforge.net/lists/listinfo/kaldi-developers
>
>