### Answer Transcription issues can stem from several factors beyond just microphone quality [(1) ]() . Here are the main solutions to improve transcription accuracy: Use Boosted Keywords - Add custom keywords (up to 100) to enhance the model's vocabulary for domain-specific terms, names, brands, or street names that might be missing from transcripts [(1) ]() [(2) ]() [(3) ]() . Adjust Denoising Mode - If you're experiencing transcription errors due to background noise or speech, experiment with the denoising mode setting [(1) ]() [(3) ]() . If background noise isn't particularly high and the agent is missing short responses like "sure" or "yes", try setting denoising mode to No Denoising (

no-denoise

) , as this preserves more of the raw audio signal and can significantly improve ASR accuracy for brief utterances [(1) ]() . Enable Transcription Mode for Accuracy - If sentences are being cut off prematurely, turn on transcription mode to optimize for accuracy . Check Your STT Mode - Ensure your speech-to-text is set appropriately. You can use

stt_mode

set to

accurate

instead of

fast

for better transcription quality [(2) ]() [(3) ]() . The denoising mode options are : -

no-denoise

- Bypass all audio denoising -

noise-cancellation

- Default setting -

noise-and-background-speech-cancellation

- More aggressive filtering Microphone quality can contribute, but

these configuration adjustments often resolve transcription issues even with standard audio quality [(1) ]() [(4) ]() .

​ To learn more, see the sources I considered: [[1] Increase transcription accuracy](https://docs.retellai.com/reliability/wrong-transcript) [[2] Create Voice Agent](https://docs.retellai.com/api-references/create-agent) [[3] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[4] Automatic Speech Recognition (ASR)](https://www.retellai.com/glossary/automatic-speech-recognition-asr) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer The issue is that your prompt is telling the agent what to say ("Let me just take another quick look") but not explicitly instructing it to wait for user confirmation before proceeding [(1) ]() . To fix this, restructure your prompt to be more explicit about the confirmation requirement:

Please be patient with the user. I can't seem to find any account for that. I got your surname as {{user_last_name}} and dob as {{user_dob}}.

When the user gives you an updated piece of information, confirm ONLY that specific piece of information back to them. For example, if they correct their date of birth, say: "Thanks for clarifying, I've updated your date of birth to [date]."

After confirming the updated information, STOP and WAIT for the user to explicitly confirm this is correct before taking any further action. Do not proceed to look up their account or take any next steps until they have confirmed the correction is accurate.

This approach addresses common interaction problems where the AI doesn't properly handle turn-taking and context . The key is being explicit about: 1. What to confirm (only the updated piece) 2. When to stop talking (after confirmation) 3. What trigger is needed to proceed (explicit user confirmation) If you're using Conversation Flow, consider using edges with transition conditions that explicitly check for user confirmation before moving to the next node [(2) ]() [(3) ]() . This gives you more deterministic control over when the agent proceeds versus relying solely on prompting.

​ To learn more, see the sources I considered: [[1] Troubleshooting AI Voice Agent Issues | Retell AI](https://www.retellai.com/blog/troubleshooting-common-issues-in-voice-agent-development) [[2] Create Conversation Flow](https://docs.retellai.com/api-references/create-conversation-flow) [[3] Subagent Node](https://docs.retellai.com/build/conversation-flow/subagent-node) ​

### Answer Your subagent prompt and transition condition need to be more explicit to ensure reliable transitions [(1) ]() . Issues with your current setup: 1. Your transition condition "the user confirms their number" is too vague - simple responses like "yes" or "ok" may not provide enough context for confident transition triggering [(2) ]() 2. Your prompt doesn't explicitly tell the agent what to do after collecting the number [(1) ]() Recommended fixes: Improved Subagent Prompt:

Your goal is to be patient and aid the user in the cancellation process.

Step 1: Gather cancellation reason
- If the user has already mentioned a reason for cancelling (e.g., "Parcel came damaged", "my GP said to cancel the nomination") , acknowledge it and move to step 2
- If no reason has been mentioned, ask: "Could you let me know the reason for your cancellation?"

Step 2: Confirm contact number
- Ask: "What's the best number if we need to get back in touch with you?"
- After the user provides their number, repeat it back: "Just to confirm, that's [number], correct?"
- WAIT for explicit confirmation (e.g., "yes", "that's right", "correct") before proceeding
- Once confirmed, STOP and prepare to transition

Improved Transition Condition: Instead of "the user confirms their number", use:

The user has explicitly confirmed their contact number is correct after you repeated it back to them

Additional tips from the debug guide : - If transitions still aren't working reliably, add transition finetune examples showing the exact user responses that should trigger the transition - Consider making your transition conditions more flexible and general if the agent struggles with short confirmations - You can also soften transition prompts to improve reliability [(2) ]() The key is being explicit about each step, what confirmation means, and when to stop talking before transitioning [(3) ]() [(1) ]() .

​ To learn more, see the sources I considered: [[1] Debug guide](https://docs.retellai.com/build/conversation-flow/debug-guide) [[2] Discord message from @daki4928 on 2026-05-09](https://discord.com/channels/1186424376300679209/1186462578310910112/1502792992581484747) [[3] Subagent Node](https://docs.retellai.com/build/conversation-flow/subagent-node) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.