So this arabic assistant started with the first msg asking if they are interested and have time. The user replied twice yeah. But the assistant didnt acknowledge that. In the recording i hear very clear audio from user. So it is clear and loud enough. And the third time the sound is picked up by the assistant which was a background sound. I dont know what is the issue. I am using a custom 11 labs voice id, using eleven-turbo v2 as TTS and GPT 4o transrciber. https://cdn.discordapp.com/attachments/1438815111883128832/1438815112424329296/Screenshot_2025-11-14_at_1.49.42_PM.png?ex=6942707e&is=69411efe&hm=7d29c08c10431248775f1b86d5279112545abe7d1c5ed44c8c2684e2e2bfc722&

Try adding "نعم" as a custom keyterm in your transcriber configuration to improve recognition of this common Arabic response. Here's how you might do it in your settings:

"wordBoost": ["نعم"]

This helps the transcriber prioritize and accurately detect "نعم" when users reply. For more details, check your dashboard's transcriber settings and test if the issue is resolved after this change (according to the documentation I have found). Source: - [Debugging voice agents](https://docs.vapi.ai/debugging)

How to add this custom keyword in transcriber configuration? Can I access these settign from Dashboard?

Message marked as helpful by @KM! 🎉

How to add this custom keyword in transcriber configuration? Can I access these settign from Dashboard?

In the dashboard open your assistant, then go to "Transcriber" and add the terms to the "Keyterms" setting. You can add many words separated by commas.

Thank you, but i cannot find the keyterms setting https://cdn.discordapp.com/attachments/1438815111883128832/1439359428489908240/image.png?ex=691a3b6d&is=6918e9ed&hm=5ce50366328c0d7aa92afbd23de7b668765c3f16065f570516eb46e576d0eb35&

Just a note, keyterms and wordboost are only available on certain transcribers like assembly-AI and deepgram. If you have a transcriber with the correct language set, you most likely wont have to configure the keyword. also we recommend using transcribers from azure, assembly-AI, gladia, and google. Use deepgram if your language is supported, then if not, try google.

Thank you. i got it

Glad you were able to get this resolved. If anything else comes up or you need further assistance, we’re here to help.