Scott Crow - 2002-11-23

Hi,

I'm attempting to hone the parameter settings for the sphinx2-phone.bat such that my recorded word "Shiny" comes out more accurately.  Currently I am getting the following results

0    51  SIL
*52  55  K
56  81  SH
82  85  EH
86  95  AY
86  107  NG
108 121  IY
*122  124 HH
125  240  SIL
241  243  SIL
*244  246  +LAUGH+
247  254   SIL

I get these results regardless of all the parameter settings I've changed and I am trying to remove the ones marked with an asterisk.  These areas are scarcely different volume-wise from the surrounding silences. 

Does anyone have any suggestions on what might be used as a threshold such that these phoneme ids might be filtered out or perhaps is there a way to derive more information for each particular phoneme that might discern the confidence of each identification?

Thanks

Scott