article

I just discovered the utterance profiler. The profiler is a nice tool. However it is not as helpful as it could be. For example, I type in an utterance and things are fine. However when I speak the same utterance, I have problems. More specifically the problem is the intent is correct but the slot values are wrong. However I see the correct slot values in the other considered intents

I suspect the lion-share of problems come during ASR. I think the utterance profiler would become a super powerful tool if

1. The considered intents had a check box. This check box would provide invaluable training information to Alexa.
2. My problems come from saying utterances, not typing them. The utterance profiler would be useful on the test page, when one is saying utterances.
3. A flag could be sent to Alexa, and the other considered intents could be sent to the back-end. This would give the back-end a fighting chance to correct stuff.

Cheers, Andrew