slow recognition times with poor recognition rates
Posted: Wed May 03, 2006 1:45 pm
I'm currently running my voiceXML app with bevocal's hosting services.
My app does a recognition based on the characters in a vehicle license plate, so yes, it is an alphanumeric grammar. While it does do a alphanumeric recognition, it is not 100% freeform... the matches are limited to the list of license plates in my database.
Bevocal uses Nuance 8.0 speech engine, and I get 99%+ recognition rates (yes, 99 times out of 100 it works flawlessly). Callers can speak the digits very slow... or so fast that the characters sound slurred together and it still works. You can speak with a regular accent, or heavily spanish-accented and it still works.
My problem is that the only times the Plum platform (running speechworks OSR 2?) recognizes my speech is when i speak..... very.... slowly.... with...... lots..... of...... pause..... between..... each..... character.
This won't fly in the real world. I need to know if there are any properties (bevocal had the "speedvsaccuracy" property) i can tweak to improve the accuracy of the recognition.
If not, do you have any plans to make a Nuance 8.0 host available on your network?
My grammar is available at: http://www.ndap.com/process/grammar-plum.php
Feel free to check it out and let me know if i've done anything wrong. I've never created a JSGF grammar before, but from what i read it looked the most similar to the .GSL grammar i had on bevocal, and converting it was very straightforward.
If I can't resolve this recognition problem, I'm not going to be able to use Plum, and that's a shame because your developer site is very good, and your access to Canada will really help us.
Thanks,
Billy
My app does a recognition based on the characters in a vehicle license plate, so yes, it is an alphanumeric grammar. While it does do a alphanumeric recognition, it is not 100% freeform... the matches are limited to the list of license plates in my database.
Bevocal uses Nuance 8.0 speech engine, and I get 99%+ recognition rates (yes, 99 times out of 100 it works flawlessly). Callers can speak the digits very slow... or so fast that the characters sound slurred together and it still works. You can speak with a regular accent, or heavily spanish-accented and it still works.
My problem is that the only times the Plum platform (running speechworks OSR 2?) recognizes my speech is when i speak..... very.... slowly.... with...... lots..... of...... pause..... between..... each..... character.
This won't fly in the real world. I need to know if there are any properties (bevocal had the "speedvsaccuracy" property) i can tweak to improve the accuracy of the recognition.
If not, do you have any plans to make a Nuance 8.0 host available on your network?
My grammar is available at: http://www.ndap.com/process/grammar-plum.php
Feel free to check it out and let me know if i've done anything wrong. I've never created a JSGF grammar before, but from what i read it looked the most similar to the .GSL grammar i had on bevocal, and converting it was very straightforward.
If I can't resolve this recognition problem, I'm not going to be able to use Plum, and that's a shame because your developer site is very good, and your access to Canada will really help us.
Thanks,
Billy