We've Moved! Please visit our new and improved forum over at our new portal: https://portal.plumvoice.com/hc/en-us/community/topics

Name Module No Match errors

Questions and answers about Plum Fuse+

Moderators: admin, support

Post Reply
cnoyes
Posts: 37
Joined: Wed Mar 01, 2017 3:03 pm

Name Module No Match errors

Post by cnoyes »

I'm having trouble getting the app to recognize my input via the Name Module. I'm in a quiet location and using simple names to test but still getting the error prompt that I'm not understood. I also tested out using the Address module to make sure something could be detected and I can only get my City/State recognized. App is Speech to Text Test. I also tried it in the sample Welcome application that was provided and I can only get City/State to be detected there, not the street address.
Can you check and see if there is something with the set up or a glitch somewhere?

thanks

cnoyes
Posts: 37
Joined: Wed Mar 01, 2017 3:03 pm

Re: Name Module No Match errors

Post by cnoyes »

Follow up question - what are the parameters with transcription and Fuse+? The names that we will be asking our users to speak will, in most cases, not be common ones. We will us two separate Name Modules to collect first and last name. When collecting the name is better to just prompt the user to spell the name rather than say it and spell it?
Is there a time constraint for that module if transcription is needed and it is not one of the common names?

support
Posts: 3632
Joined: Mon Jun 02, 2003 3:47 pm
Location: Boston, MA
Contact:

Re: Name Module No Match errors

Post by support »

We were unable to reproduce the no match errors for the address and name grammars. Unfortunately the name grammars are limited and only cover the most popular 90% of names in the US. We don't see anything wrong with your application-- voice recognition is finicky by nature. We would recommend you ask callers to enter a ZIP code using DTMF, instead of saying a city/state.

Unfortunately, transcription for name collection does not work well, like name collection over IVR does not work well in general. You could use Plum's transcription API, technical details here: http://www.plumvoice.com/docs/dev/plumd ... nscription

You should use the name module, and if a user has a nomatch error 3 times, then call the transcription API using the REST module. You can pass along the recording from the failed nomatch attempt by referencing lastresult$.recording immediately after the name collection module. You will also need to build your own webservice that can receive the transcription result. Transcription is not immediate, so your webservice that accepts transcription results would need to link that data back to your call results via transcription_id. Transcription usually takes several minutes, but could be up to 12 hours total. Also, there is no guaranteed results for transcription. If the average human can't understand what the person is saying, the transcription service won't be able to interpret either. (Right now the transcription service uses a blend of ASR and real people double checking)

We would recommend you find a workaround, like entering a unique numeric ID per doctor, or building a custom grammar that only accepts a short, finite list of known doctor's names. Open ended name collection over IVR is not a good caller experience.

Post Reply