​ To learn more, see the sources I considered: [[1] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[2] Get Voice Agent](https://docs.retellai.com/api-references/get-agent) ​

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

### Answer Based on the available sources, there's a feature that can help with this issue: normalize_for_speech. [(1) ]() [(2) ]() When you set

normalize_for_speech

to

true

, it normalizes parts of text (including punctuation, numbers, currency, dates, etc.) to their spoken form for more consistent speech synthesis. This can prevent the TTS system from incorrectly reading raw text elements like periods. To enable this feature, update your agent configuration:

json
{
  "normalize_for_speech": true
}

This setting helps ensure that punctuation and other text elements are converted to their appropriate spoken form before audio generation begins, which should reduce instances where periods are verbalized as "dot".

​ To learn more, see the sources I considered: [[1] Normalize text for speech](https://docs.retellai.com/build/normalize-text) ​

### Answer Since

normalize_for_speech

is already enabled, here are some additional troubleshooting steps: Check your language setting: The normalization feature currently supports English, Spanish, French, and German [(1) ]() . If you're using a different language, it will result in a no-op (text not modified) . Verify your agent's language configuration matches one of these supported languages. Review your prompt/response generation: The issue might be occurring at the LLM level before normalization happens. Check if your GPT-4 model is generating responses with excessive periods or unusual punctuation patterns. You may need to adjust your system prompt to discourage unnecessary punctuation. Consider TTS provider behavior: The normalization adds approximately 100ms of latency to the process . If you're using a specific TTS provider, they may have their own handling of punctuation that conflicts with or bypasses the normalization step. Test with different voices: Try switching to a different TTS voice to see if the issue persists, as some voices may interpret punctuation differently. If the issue continues after checking these items, I'd recommend reaching out through the [enterprise plan contact]() for more specific debugging assistance with your agent configuration.

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

Thanks for the follow-up, @K Rudy. I'll get back to you soon.

### Answer Based on the available sources, both

gpt-5.4

and

gpt-5.4-mini

are listed as available model options for Retell AI agents [(1) ]() [(2) ]() [(3) ]() [(4) ]() [(5) ]() . However, the sources don't provide specific details about the differences between these two models. Typically, "mini" versions of models are optimized for faster response times and lower costs, while standard versions offer more advanced reasoning capabilities. But I don't have documentation that explicitly describes the differences between GPT-5.4 and GPT-5.4-mini in the Retell AI context. For detailed information about the specific differences between these models and which would be best for your use case, I'd recommend: - Visiting the [Documentation]() - Visiting [Hire certified Retell partner]() Would you like a general answer based on my knowledge instead?

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

### Answer Based on the available information, here's what differentiates GPT-5.4 from GPT-5.4-mini: GPT-5.4 is designed for complex workflows requiring higher intelligence, while GPT-5.4-mini is optimized for lightweight, high-speed tasks [(1) ]() . Key differences: **Intelligence**: GPT-5.4-mini offers the highest intelligence in the lightweight model group, while GPT-5.4 nano provides high intelligence at a medium level . **Latency**: GPT-5.4-mini has medium latency (around 800ms) , whereas GPT-5.4 nano shows medium latency (around 700ms) . **Pricing**: GPT-5.4-mini costs $0.012/min, while GPT-5.4 nano is $0.003/min . **Use cases**: - GPT-5.4-mini is recommended for small tasks that still benefit from reasoning capabilities - GPT-5.4 nano is best for cost-sensitive businesses that don't require reasoning or latency boosts The "mini" variant balances strong intelligence with reasonable cost for real-time agents, making it a good middle ground .

​ To learn more, see the sources I considered: [[1] Retell AI Launches GPT-5 for Voice Agents – Higher Intelligence, Smarter Conversations | Retell AI](https://www.retellai.com/blog/gpt-5-now-on-retell-smarter-ai-voice-agents-with-reasoning-power) ​

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