Welcome to GotSpeech.NET Sign in | Join | Help

Marshall Harrison - "the gotspeech guy"

Site news, Speech Server insight and assorted ramblings
Write Mr. Wright with the right directions.

Write Mr. Wright with the right directions.

The above sentence is a classic problem for speech recognition (and dictation). All three words are pronounced the same (by most people) but have different meanings depending on their usage in the sentence.

"Invalid" is another word that has similar problems (actually the English languages has a lot of problem words). It can mean that something is not valid or it can mean handicapped or hindered in some way. Luckily for us it is pronounced differently in each case but it could cause problems when using TTS.

I mentioned this to point out that the people in MIT's Media Lab are working on something called The Open Mind Common Sense Project that is building a vocabulary database. The database associates words with phrases to find the best match.

This is interesting stuff. Read the article and let me know what you think.

Posted: Monday, May 08, 2006 8:19 AM by marshallharrison

Comments

IanRae said:

Another good one is "Dr. Smith lives on Maple Dr." which MSS correctly pronounces as "Doctor Smith lives on Maple Drive.".

There's a Tips and Tricks entry on this at:
http://www.microsoft.com/speech/techinfo/tipsandtricks/default.mspx
# May 9, 2006 10:05 AM

marshallharrison said:

Which tip are you referring to Ian?
# May 9, 2006 1:32 PM

IanRae said:

The question under tips & tricks is: Q. How does the speech recognition engine in Microsoft Speech Server 2004 treat abbreviations, digit strings, dollar amounts, etc?

There is a table of abbreviations (such as "oz" pronounced as "ounce").  Doctor and Drive for Dr is mentioned as a context-sensitive case.

# May 10, 2006 2:57 PM

marshallharrison said:

It does do pretty well with abbreviations.

I need to write something to see how it does with "write", "right" and "Wright"
# May 10, 2006 9:38 PM
Anonymous comments are disabled