Automatic Speech Recognition (ASR) Oracle Database Tips by Donald Burleson

Often a new technology is driven by necessity.  For example, a quadriplegic programmer has developed an automatic speech recognition system entirely out of necessity because he does not have the use of his hands, and the only thing that had been holding him back from making commercially available products was the high cost of RAM memory.  At the time it required about a hundred megabytes of memory, but now he's got what they call continuous speech recognition capability with 200 meg of RAM, which is still expensive for today's PC's, but is certainly not out of reach for many different kinds of applications. 

This product has become a commercial product for use by physicians and attorneys, where they can just go directly to a machine and dictate their information directly onto the computer without having to have people transcribe that kind of information.  So again, here, within the marketplace, clerical skills are going to be going away as we know them.  Just as within the last twenty years, nobody learns Gregg shorthand in secretarial school anymore because it's been displaced by tape recorders, so will become keyboarding is going to be very obsolete.

A lot of the new technology has been driven by the U.S. Federal Government.  If we examine automatic speech recognition (ASR) it started with the CIA doing research for the Northwest Parallel Computing Resource Center in Syracuse, New York.  The Government  has been working on databases where a user can make an English query into a Russian database and return the results said in French.  These kinds of dynamic language translations are of immense value but they didn't come without a cost.  In the 1980s IBM spent years and millions of dollars learning how to parse spoken English with a computer.  For example, it is very difficult for a computer to understand that the word "bus station" is two words and not one, since they are pronounced as a single sound. 

But parsing out the spoken words is just the beginning.  Once you've then parsed the word, you still have to be able to derive meaning from the spoken word, and that's a very difficult challenge in and of itself, especially because of the natural ambiguities that come out of the human language.  Just a simple phrase, "Mary had a little lamb", computer will get confused and it can't resolve the verb "had" in the sentence.  The computer would ask "What do you mean, Mary had a little lamb?  Did Mary buy a little lamb?  Did Mary eat a little lamb?  Do we take a biblical interpretation of "had"?  What do you mean Mary "had" a little lamb"?

These kinds of natural ambiguities occur in human English all the time and it's very difficult to get computers to resolve the nuances of the English language.  For example, there was a product in the 1970s marketed by Excalibur Corporation called Savvy, which was a natural language interface to databases.  It worked for very simple queries, but when you gave it something sophisticated, or ambiguous like "How long has Joe been with us?", Savvy would come back and say, "What are you asking for, Joe's date of birth?, or Joe's date of hire?". 

See, so you have these kinds of natural ambiguities that have to be addressed, and the language translation routines often produce hilarious results.  Consider the phrase, "the spirit is willing, but the flesh is weak", it was translated it into Russian, then the translation was again translated back into English and it stated, "the vodka is good, but the meat is rotten".  Hardly a literal translation of that phrase. 

But this type of technology is getting much more sophisticated and we are actually going to start seeing ASR, the whole area of Automatic Speech Recognition, taking over the desktops.  We will actually be keyboard and mouse free, and people with actually interact with their computers, much as they would if they  were individuals today.  Within the more ho hum, mundane realm of information systems, we're finally seeing another nascent technology finally taking off.  And here, we're talking about object technology.

 

---