When working on improving our product, we'd like to capture field failures that could be used to train our custom speech-to-text model. We have a wake word (handled locally) and then we establish a connection to IBM Watson's Speech-to-Text service for a more complicated command. If the command was not understood, there are four possibilities:
1) User error - invalid or no command
2) User changed his mind after saying wake word - no command
3) Wake Word detection error - the wake word heard but not actually said, so no command was issued
4) Command detections error - Watson didn't correctly understand what the user said
Being able to retrieve from Watson, after the fact, interesting sound files could help us find things that fall into case #4 and help us train our custom model better.
|Who would benefit from this IDEA?||Customers who want to train custom voice models|
NOTICE TO EU RESIDENTS: per EU Data Protection Policy, if you wish to remove your personal information from the IBM ideas portal, please login to the ideas portal using your previously registered information then change your email to "firstname.lastname@example.org" and first name to "anonymous" and last name to "anonymous". This will ensure that IBM will not send any emails to you about all idea submissions