Could someone let me know whether the Insertion error, Deletition and
Substitution error are ONLy applicable to Word recognition and NOT to
continuous SR.
I wonder what is the exact definition for those errors.
Thanks
Sent via Deja.com http://www.deja.com/
Before you buy.
In simple terms: An insertion error is when a word is recognized when
none was spoken. A deletion error is when no word is recognized when one
was spoken. And a substitution error is when the wrong word was
recognized.
I take it that with "Word recognition" you mean single word recognition?
In that case, if you are only looking at the recognizer itself there are
no insertion or deletion errors. This is because the user utters one
word and the recognizer outputs one word. The only possible error in
this case is that the recognizer outputs the wrong word. This is called
a substitution error.
It is a bit different if you also evaluate the segmentation. In that
case you get an insertion error if the segmentation detects an utterance
and the recognizer outputs a word even though the user didn't say
anything. Conversely, you get a deletion error if the user says
something but the segmentation doesn't detect it.
You can also have deletion errors if the recognizer has some kind of
confidence threshold. In that case even if the segmentation correctly
detected the user utterance, the recognizer might not output anything if
the confidence is too low, resulting in a deletion error.
For continuous speech recognition there are additional possibilities: If
the user utters an entire sentence then the recognizer result might
contain additional words which were not spoken. These are insertion
errors. Or it might not contain some words, which would be a deletion
error. Of course, words might also be misrecognized, which is a
substitution error.
Example: The user says "It's a nice day at the beach" and the recognizer
outputs "It's nice at in the breach". In this case there are two
deletion, one insertion, and one substitution errors.
Hope this helps,
Marcus
--
Dr. Marcus E. Hennecke
TEMIC Telefunken Microelectronic GmbH
Speech Processing Tel +49-731-3994-117
Soeflinger Str. 100 Fax +49-731-3994-250