### Answer There isn't a specific "guardrail" feature documented for handling transcription errors in Retell AI. However, there are several settings and strategies you can use to improve transcription accuracy and mitigate errors: ## Improving Transcription Accuracy Boosted Keywords – Add custom keywords to enhance the model's vocabulary for domain-specific terms (up to 100 keywords) . [(1) ]() Denoising Mode – Play with the denoising mode setting to filter out background noise or unwanted speech. Options include

noise-cancellation

and

noise-and-background-speech-cancellation

. [(2) ]() [(1) ]() STT Mode – Switch from

fast

to

accurate

mode if transcription is being cut off prematurely. ## Configuration Options via API You can configure these settings when updating your agent: [(2) ]() -

boosted_keywords

– List of words to bias the transcriber toward -

denoising_mode

– Options:

none

,

noise-cancellation

,

noise-and-background-speech-cancellation

-

stt_mode

– Options:

fast

,

accurate

,

custom

## Workarounds for Input Errors For capturing sensitive data like phone numbers or emails where transcription errors are common, consider using *User Keypad Input (DTMF) * to let callers enter numbers via keypad instead of voice. [(3) ]() The sources don't document a built-in "guardrail" mechanism that automatically detects and handles transcription errors. If you need specific error-handling logic, you may need to implement validation in your conversation flow or external webhook.

​ To learn more, see the sources I considered: [[1] Increase transcription accuracy](https://docs.retellai.com/reliability/wrong-transcript) [[2] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[3] Platform Changelogs | Retell AI](https://www.retellai.com/changelog) ​

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

### Answer Based on the available sources, I don't have specific prompt examples that users add to mitigate transcription errors. The documentation focuses on configuration-level solutions rather than prompt-based workarounds. ## What the Sources Do Mention For language issues, one community suggestion was to add explicit language instructions in the prompt: [(1) ]()

# Response Guidelines
* You MUST speak only in Hindi

For company/name transcription errors, the recommended approach is using Boosted Keywords to bias the speech-to-text model toward specific terms rather than prompt engineering. [(2) ]() ## Recommended Configuration Approaches The official documentation emphasizes these methods over prompt modifications: [(3) ]() - Boosted Keywords – Add domain-specific terms (up to 100 keywords) - Denoising Mode – Filter background noise/speech - Transcription Mode – Switch to accuracy-optimized mode if sentences are being cut off If you're looking for prompt-based strategies that other users have implemented, I'd suggest asking in the community or checking: - [Increase transcription accuracy]() - [Discord]() Would you like a general answer about common prompt strategies for handling transcription uncertainty?

​ To learn more, see the sources I considered: [[1] Facing Issue in languages](https://community.retellai.com/t/facing-issue-in-languages/384) [[2] Discord message from @erika_jane9 on 2024-09-11](https://discord.com/channels/1186424376300679209/1186462578310910112/1283435022405079153) [[3] Increase transcription accuracy](https://docs.retellai.com/reliability/wrong-transcript) ​

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