On Wed, Jun 11, 2008 at 9:17 PM, Heiga ZEN (Byung Ha CHUN) <
zen@xxxxxxxxxxxxxxx> wrote:
Hi,
QiXiong Tu wrote (2008/06/11 17:49):
When generating a regression tree, I want to specify *-aa+* and *-aw+* into two different classes. The RC command in HHed is as follow: RC 32 "reg" [itemlist]
I use [itemlist] = {*-aa+*.state[3].stream[1-4],*-aw+*.state[3].stream[1-4]}, but it doesn't work.
How can I build two trees for the *-aa+* and *-aw+*,respectively?
In HTKBook section 17.8.1, you can find the following statements:
RC N identifier [itemlist]
It is also possible to specify an itemlist containing the "non-speech" sound components such as the silence mixture components. If this is included then the first split made will result in one leaf containing
the specified non-speech sound components, while the other leaf will contain the rest of the model set components. Tree construction then continues as usual.
As described above, [itemlist] is used to specify non-speech models, which will be divided at the first split.
Therefore,
RC 32 "reg" {*-aa+*.state[3].stream[1-4],*-aw+*.state[3].stream[1-4]}
divides {*-aa+*.state[3].stream[1-4],*-aw+*.state[3].stream[1-4]} and others at
the first split.
If your model set has only *-aa+* and *-aw+*,
RC 32 "reg" {*-aa+*.state[2-4].stream[1-4]}
will split *-aa+* and *-aw+* at the first split.
Otherwise, building two regression trees for *-aa+* and *-aw+* is not supported in HTS.
You need to modify HHEd.c by yourself.
Regards,
Heiga ZEN (Byung Ha CHUN)
--
------------------------------------------------
Heiga ZEN (in Japanese pronunciation)
Byung Ha CHUN (in Korean pronunciation)
Department of Computer Science and Engineering
Nagoya Institute of Technology
Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan
http://www.sp.nitech.ac.jp/~zen
------------------------------------------------