Acoustic Model Discussions

Flat
HHEd [Error 2662]
User: samsao
Date: 1/18/2008 2:08 am
Views: 7664
Rating: 30

Hi,

 I am running a  Chinese(Mandarin)  LVCSR, but I came across the error below, while I am running the below clusterring command:

HHEd -A -D -V -B -w clustering/hmm10_1000_800/MMF -d clustering/hmm1 -C cfgs/hhed-cluster.cfg -M\    clustering/hmm10_1000_800 -H clustering/hmm1/MMF clustering/hmm10_1000_800/cluster_1000_800.hed train.xwrd.mlist
=============Error=================== 

ERROR [+2662]  AssignStructure: cannot find tree for ei1-uai3+ue4 state 2

==================================

  1. The phones used in this system are tonal phonemes
  2. The tree.hed file contains command lines below:

................... 

QS "L_zh" { zh-* }
QS "R_zh" { *+zh }
TR 1
RO ROVAL "clustering/hmm1/stats"
TR 2
TB TBVAL "a_s2_" { ( a1, a2, a3, a4, a5, *-a1, *-a2, *-a3, *-a4, *-a5, a1+*, a2+*, a3+*, a4+*, a5+*, *-a1+*, *-a2+*, *-a3+*, *-a4+*, *-a5+* ).state[2] }
TB TBVAL "a_s3_" { ( a1, a2, a3, a4, a5, *-a1, *-a2, *-a3, *-a4, *-a5, a1+*, a2+*, a3+*, a4+*, a5+*, *-a1+*, *-a2+*, *-a3+*, *-a4+*, *-a5+* ).state[3] }

......................

     3. The "unseen" file has all possible tri-groups of tonal phonemes 

a1-a1+a1
a1-a1+a2
a1-a1+a3
a1-a1+a4
a1-a1+a5
a1-a1+ai1
a1-a1+ai2
a1-a1+ai3
a1-a1+ai4
a1-a1+ai5
a1-a1+ao1
a1-a1+ao2
............

 

Thank you. Any help would be appreciated

--- (Edited on 1/18/2008 2:08 am [GMT-0600] by samsao) ---

--- (Edited on 1/18/2008 2:16 am [GMT-0600] by samsao) ---

Re: HHEd [Error 2662]
User: kmaclean
Date: 1/18/2008 3:34 pm
Views: 264
Rating: 21

Hi samsoa,

I'm not really sure, but there are some hints in the HTK book and the email archives that might help you: 

>=============Error=================== 

>ERROR [+2662]  AssignStructure: cannot find tree for ei1-uai3+ue4 state 2

>==================================

The HTK Book says:

+2662 Trying to sythesize for unknown model
      There is no tree or prototype model for the new context. Ensure that a tree has been constructed for the base phone.

I am assuming that the base phone in this case is  "uai3" - does it exist for state 2?

From the HTK email archives (you need to be logged in) someone got a similar error:

 > ...............
> LT "tree1"
> AU "/daten/lars/listen/vocab/worte.vocab.triphone"
> CO "triphoneX.tied"
> ................
>
> The script above produces the following error message.
>
> ERROR [+2662]  AssignStructure: cannot find tree for U-N+aU state 2

Gunnar Evermann (12 Apr 2001) replied:

AU assumes that you have a tree for every state.

In another email, someone had wrote this:

> I try to run this command:
> HHEd -B -H hmm12/macros -H hmm12/hmmdefs -M hmm13
> tree.hed triphones1 > log
> but it prints out an error message:
> "ERROR [+2662]  AssignStructure: cannot find tree for b state 2"

Yifan Zhang said (Tue, 6 Sep 2005):

Check if there is a tree for b. You should have a commandline like TB 350.00 b_s2 {b-*,....} If you do have this line, check the first item in the itemlist and makesure it's basephone is b

Ken 

--- (Edited on 1/18/2008 4:34 pm [GMT-0500] by kmaclean) ---

Re: HHEd [Error 2662]
User: samsao
Date: 1/18/2008 7:02 pm
Views: 3088
Rating: 31

Hi, Ken

 Thanks so much for ur quick reply 

--- (Edited on 1/18/2008 7:02 pm [GMT-0600] by samsao) ---

PreviousNext