Page 1 of 1

Capturing US City and State Value

Posted: Wed Aug 25, 2004 5:47 pm
by mnicolls
Another VoiceXML hosting company I have worked with offers a built-in grammar for capturing US City and State. Is this possible with Plum? If so, what about cities in Canada and Mexico?

IVR system does not support all types of built-in grammar

Posted: Thu Aug 26, 2004 10:01 am
by support
You could build that grammar from scratch, but Plum IVR system does not yet support that kind of built-in IVR grammar.

Hope this helps, please feel free to post back addtional questions :)

Plum Support

Posted: Thu Aug 26, 2004 10:10 am
by mnicolls
support wrote:You could build that grammar from scratch...
I considered that, but our city database has 60,000+ records so the grammar would be huge. This raises other questions:
  • 1) Would this create performance issues?
    2) Does the grammar parser cache or precompile grammars?
    3)How it will work with odd city names, like "Des Peres", which is pronounced "Da Pear"; would we have to create pheonetic spellings of the city names to accomodate this?
    4) Has someone already done this?

fix IVR issue by partitioning data into managable partitions

Posted: Thu Aug 26, 2004 10:26 am
by support
1) Yes the IVR grammar parser will cache compiled IVR grammars

2) Generally speaking, it doesn't handle odd city names. You would likely have to provide pronuciations for each name as well as variants for each of the ways it could be pronounced.

3) For convenience, maybe the top 12 most popular are selectable. One way you could fix this IVR issue is to partition the data into more managable partitions. For example, requiring a prefecture or state would significantly reduce the set of available choices.

Hope this helps!
Plum Support

Posted: Thu Aug 26, 2004 10:33 am
by mnicolls
Thanks for the quick response! I wonder if there is a way to write a grammar to take in the utterances, and only evaluate the state, then pass the users utterances to another grammar to resolve the city.

For example, the user might say "Saint Louis, Missouri". The grammar could resolve “Missouri” against a grammar with only the 50 states. Then it might pass on the rest of the utterance to a grammar that contains only the city names for Missouri.

This would prevent having to prompt the user for state, then city. IT is much more natural to say "Saint Louis, Missouri" than to say "Missouri", then wait for another prompt, then say "Saint Louis".

This must be possible, because BeVocal offers this as a built-in grammar, and I haven't found a US city that it doesn't recognize! I think they said it was a Nuance grammar - does that mean that there is an existing grammar out there that you can download/purchase from Nuance?

-M

city/state grammars to be released for IVR grammars

Posted: Fri Aug 27, 2004 10:04 am
by support
you could write more than one IVR grammar and then link them together, but you can't chain multiple asr attempts together; IVR grammars just don't work that way.

the big issue is with how people pronounce things... you can't just have a city state list, you would likely have 10 pronunciations for every city and state.

We are working on city/state IVR grammars to be supported internally, but for the immediate future the feature is not available...we will be adding it in a future IVR release at some point.

For now, we are not supporting Nuance as a Speech Rec engine, but we are considering it; this is still in development so it won't be available for a while if and when we decide to officially roll this into production. IBM Speechworks is available for our customers if they opt for it.

Hope this is of help to you!

support