I know from context-independent to context-dependent scheme, there will be too much state leading to insufficient training. so we need to using clustering technique to reduce state or tie state. My question is how to control this process, for example, I don't want to tie state, because my current phonetic inventory size is very small. so I don't want to tie my current triphone.
Thanks in advance!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
You could try setting the num_leaves to the cube of the number of
phones in your inventory, and turn off the clustering (I believe
setting cluster-thresh to zero does this, but I forget where in the
script this is to be done). If you have sparse data, though, it might
still do some tying.
Dan
I know from context-independent to context-dependent scheme, there will be
too much state leading to insufficient training. so we need to using
clustering technique to reduce state or tie state. My question is how to
control this process, for example, I don't want to tie state, because my
current phonetic inventory size is very small. so I don't want to tie my
current triphone.
Dear all,
I know from context-independent to context-dependent scheme, there will be too much state leading to insufficient training. so we need to using clustering technique to reduce state or tie state. My question is how to control this process, for example, I don't want to tie state, because my current phonetic inventory size is very small. so I don't want to tie my current triphone.
Thanks in advance!
You could try setting the num_leaves to the cube of the number of
phones in your inventory, and turn off the clustering (I believe
setting cluster-thresh to zero does this, but I forget where in the
script this is to be done). If you have sparse data, though, it might
still do some tying.
Dan
On Sun, Jul 12, 2015 at 4:45 PM, blculiwei blculiwei@users.sf.net wrote: