Voiceflow, is matching my utterance respond to the wrong intents

I’m building an Alexa app (Soooo hard by the way :sob:).

I have at least 10 flows that contains at least 30 blocks each.

Out of my many choice blocks, for some reason my utterance from another choice block (intent) will match with an intent from another choice block! Remind you these two intents have nothing in common.

This phenomena also occurs with smaller voiceflow projects I’ve created.

I feel like this is a very common issue, and probably been solved many times. But I can’t find the topic. LOL!

But if anyone can help I really really appreciate it!

could you give us some screenshots of your flow?

Here’s a case in which Alexa is requesting a date from the user. The expected response can either be an actual date or number of days (example 5 days).

Issue#1: Activates wrong intent when user enters an actual number. This is suppose to activate “get_invoice_term_days” not “get_invoice_term_date”.

The intent “get_invoice_term_days”, has 3 utterance in it. The first only expects a number.


issue#2: When activating utterance 2 or 3 , it once again activates a totally different intent. See the images.





For your reference, here’s both intents again.
get_invoice_term_date

I tested a small sample in my end with my locale(ja-JP), it works. I don’t know the difference…

Anyway,

  • for Alexa projects, you should test on Alexa Developer Console. Voiceflow’s test does not simulate Alexa perfectly (because Amazon is not open, I guess.)
  • Also, it is important to test with voice on ADC, not type. Alexa converts voice to text and sometimes it is different from what you expect.

Since days and dates as utterances have overlapping sounds, e.g. 19th and 19 days, Alexa’s NLP brain may be having trouble routing your utterance to your desired intent. In my own experience, any phoenemes that overlap, e.g. got, bought, daughter - have the potential to confuse Alexa (there are other factors at play too, e.g. poor pronunciation by user).

To test this theory, in your case, you may want to split your prompt into separate steps to separate out the parsing jobs by Alexa. Plus, this approach might reduce the cognitive demand on the user, e.g.


Alexa: How many days does your client have to pay the invoice?

User: What? How would I know that?

Alexa: No worries, can you tell me the due date of the payment?

User: Oh, that’s December 14, 2022

Alexa: Just to confirm, you said December 14, 2022, yes?

User: Yep.

Alexa: Got it. Now, let’s verify the project information …


Terry