[Tcl/SMAPI-discuss] Meaning of Confidence Score

[Sunil V. <ve...@me...>]

I'm writing some code to capture the confidence score during
recognition.  I'm unable to determine the meaning of the numbers I'm
getting.  I've searched the docs, via voice dev site, discussion forums,
and the web for more information on this.  The only thing I can find is:

> http://www-4.ibm.com/software/speech/dev/faq_windows.html
> 
> What is Phrase Score?
> 
> In brief, the score is not a confidence one, it is an average acoustic score per second. The acoustic score depends on the quality of the match and the length of the speech aligned with the word ... usually longer words get higher scores. There is no clamping of the range ... negative values are possible, as are values above 100. ISVs will have to experiment to determine what are reasonable/unreasonable values for the particular words in there applications.
> 


However, this still doesn't give an indication how to interpret the
number.  Ideally, I'd like a probability (range from 0.0-1.0).  Is there
a way to convert these scores?  Is there a different call I should be
using?


FYI: the relevant code segment I'm using to get the score is:

smCallback recognizedText { 
  foreach w [smWord -spell -startTime -endTime -score [getFirmWords]] {
    puts [format "%s\t%d\t%d\t%d" [lindex $w 0] [expr [lindex $w 1] -
$startTime] [expr [lindex $w 2]-$startTime] [lindex $w 3]]
  }
}

and it produces output which looks like:

are     2983    3163    -5
the     3582    3652    -6
best    3652    4031    -11
this    5108    5278    -10
is      5278    5388    0
tied    5388    5597    -16
to      5597    5657    -7
avoid   5657    6086    -14
have    12621   12781   -17


Many thanks,
--Sunil