Genesys Cloud - Main

Hi team,

HNY 2026 to all! 

We have been looking at our transcription accuracy and have found it tricky that the dictionary feature is limited to only allowing 10 sounds like entries per dictionary term. 

We have experienced multiple variations for some terms which throws off transcription accuracy.

It seems a bit limiting to expect that some terms won't ever be perfect because of the limitation of 10 sounds like phrases per term. How have other customers gotten around this? 

Also thinking it would require significant analysis effort to work out which ones are higher volume errors. Are there any tips to speed this up?

Cheers

#Reporting/Analytics
#WEM-Quality,WFM,Gamification,etc

------------------------------
Deepa Galaiya
Product Owner, Customer Interactions
------------------------------

Hello Deepa, 

Happy New year to you and yours as well!

I want to break down everything you brought up and hopefully give you some helpful approaches. Lets start with efficiently identifying and prioritizing which terms need attention. I would start by focusing on reviewing at least two hours of randomly selected recordings across your org. Then implementing a crowdsourcing approach to log commonly mis-transcribed words (by crowdsourcing I mean supervisors and agents within your org). This should help speed up and identify high-volume errors without requiring extensive individual analysis. 

Next lets move on to the 10 sounds like limitation. I would star with using the sensitivity boost values through the API to enhance recognition of critical terms. I would then try to focus on the most common variations rather than trying to capture every possible variation. Also consider grouping similar variations under broader terms. 

Hope this helps!

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

Hi,

We've faced the exact same challenge on our side. In a few use cases, having only 10 "sounds like" entries per dictionary term is not enough, especially when there are many phonetic variations that directly impact transcription accuracy.

We even opened a support ticket with Genesys to check whether this limit could be increased, but unfortunately today this is a hard platform limitation and cannot be adjusted.

As a result, some level of imperfection is almost unavoidable when the number of real-world variations exceeds that cap.

That said, there is already an Ideas Portal request asking for an increase to the "sounds like" limit:
👉 https://genesyscloud.ideas.aha.io/ideas/DARSTA-I-332

If this is something impacting your operation, I'd strongly recommend voting on that idea. The more traction it gets, the higher the chance it's prioritized by the product team.

In the meantime, what has helped us is focusing the limited slots on the highest-impact variations (those that most affect downstream analytics or automation), even though it does require some upfront analysis effort.

Hope this helps - and fingers crossed this idea moves forward soon.

Cheers!

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