Hello, I created a Phone Agent using a Twilio account I have access to, but the user experience has - Thread Discussion

# 🛠technical-support

Julian·12 replies

Hello, I created a Phone Agent using a Twilio account I have access to, but the user experience has been quite poor. Here are the main issues I encountered:

The Spanish voice sounds like it's from a different country, which negatively impacts the user experience. Unfortunately, there's no way to change or customize the voice or accent.

The agent doesn’t follow the directives defined in the prompt, which limits control over the conversation.

Most of the time, when I call the Phone Agent, it only plays the welcome message and then goes silent. This behavior is inconsistent with how it performs in the yourGPT console.

Occasionally, I hear keyboard clicks or background voices in English during the call, which is confusing and unprofessional.

Overall, it's been a very frustrating experience.

Rohit | YourGPT18/06/2025 14:27

Hello @Julian,

The feature is currently in beta, and we’d be happy to hear your feedback and collaborate closely on your use case.

Let review you voice agent quickly, I will he happy to help you setup.

Rohit | YourGPT18/06/2025 14:28

@Julian Can you please confirm on which Chatbot you have setup the Phone Agent?

Julian18/06/2025 14:30

Hello @Rohit Joshi , I received an email saying "📞 Phone Voice Agents

Your AI agent can now make and receive phone calls, handling customer queries automatically. Escalate to a human agent any time when needed." that is why I expended time. Now I'm exploring another solution with native language like https://www.cognigy.com/ . The agent I was trying to use with this is Cerebro

Rohit | YourGPT18/06/2025 14:32

okay, please give me few minutes to take a look

Rohit | YourGPT18/06/2025 14:34

As I have just checked, I found that you are using GPT4.1 model, which is not a realtime model.

have you tested GPT4o realtime or GPT4o mini realtime?

Rohit | YourGPT18/06/2025 14:37

@Julian GPT-4o & GPT-4o realtime for voice offers real-time, natural conversations with ultra-low latency, native speech input/output (no separate ASR or TTS needed), emotional and multilingual understanding, and the ability to speak and listen simultaneously—making it ideal for voice agents.

Rohit | YourGPT18/06/2025 14:37

"The Spanish voice sounds like it's from a different country, which negatively impacts the user experience. Unfortunately, there's no way to change or customize the voice or accent."

can you test this with GPT4o?

Julian18/06/2025 14:37

yes, same results

Rohit | YourGPT18/06/2025 14:38

have you added anything in instructions for the accent?

Rohit | YourGPT19/06/2025 16:51

hello @Julian,

I have an update for you.

We’ve recently added support for Google Gemini Flash 2.0. Could you please test this model and let us know if you find it better?

Kindly start by testing in English first, and then try it in other languages as well.

We’ve tested it in English & Hindi and found the results to be really amazing.

Julian19/06/2025 17:25

Hello Rohit, thank you, I will test that today and then I’ll send you my feedback. Thanks!

Julian19/06/2025 19:24

"Rohit, I tried again, but got the same result. The user experience is actually worse with the Spanish voice — it sounds like a non-native speaker (a 'gringo') trying to speak Spanish."