Genesys Cloud - Main

 View Only

Sign Up

  • 1.  Bot recognition issue

    Posted 01-14-2026 03:36

    The bot is incorrectly recognizing "Year" as "Ear" or "Yeah."
    ASR intent confidence is 0.5,
    but classification fails by matching too many incorrect inputs. Still, if someone can suggest something else to avoid this failure, it would be great.


    #CommunityQuestions(Contest,Community,etc.)

    ------------------------------
    Anand
    NA
    ------------------------------


  • 2.  RE: Bot recognition issue

    Posted 01-14-2026 13:50

    Hello Anand,

    I have a few ideas that may help.

    The first being using a traditional grammar-based ASR approach, this should provide a higher accuracy for constrained domains. Create specific GRXML grammar files to define expected inputs more precisely. 

    The second, would be grammar optimization. Adding weights and probabilities to your grammar rules to favor "Year" over similar-sounding words. This would include context specific rules that make "Year" more likely in relevant scenarios and remove ambiguity by defining clear contexts where "Year" would be expected. 

    Technical recommendations would be to increase the confidence threshold for recognition if 0.5 is resulting in too many false positives and implement slot-level confidence scoring to better validate the recognition. 

    Hope this helps!



    ------------------------------
    Cameron
    Online Community Manager/Moderator
    ------------------------------

    Original Message


  • 3.  RE: Bot recognition issue
    Best Answer

    Posted 01-14-2026 14:42
      |   view attached

    One suggestion would be to test the different Bot / ASR engines available in the Bot Flow (Default, Genesys Enhanced v1, v2, v3).

    There isn't a single "best" engine that works for all organizations. In practice, each engine can behave differently depending on the org, language, accents, and specific use cases, so it's not possible to say in advance which one will perform better.

    The safest approach is to:

    • Change the default engine for the language

    • Run controlled tests using the same utterances

    • Compare transcription accuracy, intent confidence, and misclassification rates

    Only through testing can you identify which engine works best for your scenario.

    If this answer helps, please consider marking it as the best answer so others facing similar ASR challenges can benefit as well.



    ------------------------------
    Mateus Nunes
    Tech Leader Of CX at Solve4ME
    Brazil
    ------------------------------

    Original Message


  • 4.  RE: Bot recognition issue

    Posted 3 days ago

    I am experiencing the same issue. Thank you for sharing your suggestion, @Mateus Nunes.



    ------------------------------
    Srikanth k s
    ------------------------------

    Original Message


Related Content