Speech rec works. Talk to Sears, TellMe and Schwab to name a few
This article is not adressing John Doe controlling his PDA with speech commands(although we're getting there). It's talking of Voice browsing the way you browse the web.
We get very high recognition acuracy (90-99% range). Heck, sometimes I listen to.wav where I don't have a clue what the user is saying and the engine is righ on! And with good design you'd be amazed at what voice portals can accomplish.
The trick with voice applications is to design excellent dialogs. Basically, the speech rec will be good if you know what the user is going to say. That means asking the right questions to get the right answers. Then write the grammars that matches. System that are speaker-independant need to have a grammar specified. Basically a list of possible answers the user can give. This network of possibilety can be small or huge (think of diallers with 100k names)
Text-To-Speech also has a bad reputation. We all remember what Dr. SBAITSO used to sound like but beleive me, when the next generation of TTS engine will speak to you you probably won't notice. I heard samples from a few engines and they are unbeleivable.
Take a look at the Voice Web Server. It is available for free (free registration required) on extranet.nuance.com and it works on any Intel PC (yeah, I know...) with a SB16 compatible soundcard
Then come discuss with other users on our discussion forums. I'll be there to answer your questions
Slashdot Mirror
Disclaimer: I work for Nuance Communications a speech recognition software company
.wav where I don't have a clue what the user is saying and the engine is righ on! And with good design you'd be amazed at what voice portals can accomplish.
Speech rec works. Talk to Sears, TellMe and Schwab to name a few
This article is not adressing John Doe controlling his PDA with speech commands(although we're getting there). It's talking of Voice browsing the way you browse the web.
We get very high recognition acuracy (90-99% range). Heck, sometimes I listen to
The trick with voice applications is to design excellent dialogs. Basically, the speech rec will be good if you know what the user is going to say. That means asking the right questions to get the right answers. Then write the grammars that matches. System that are speaker-independant need to have a grammar specified. Basically a list of possible answers the user can give. This network of possibilety can be small or huge (think of diallers with 100k names)
Text-To-Speech also has a bad reputation. We all remember what Dr. SBAITSO used to sound like but beleive me, when the next generation of TTS engine will speak to you you probably won't notice. I heard samples from a few engines and they are unbeleivable.
Take a look at the Voice Web Server. It is available for free (free registration required) on extranet.nuance.com and it works on any Intel PC (yeah, I know...) with a SB16 compatible soundcard
Then come discuss with other users on our discussion forums. I'll be there to answer your questions
Mathieu