Custom Voice Not Triggering Voice Generation Request for First Message Vapi AI #support

Custom Voice Not Triggering Voice Generation Reque...

Kfir Ochana

06/29/2025, 10:55 PM

## Title: Custom Voice Not Triggering Voice Generation Request for First Message Issue Summary: When using a custom voice configuration, the agent fails to send a voice generation request for the first message, even when

firstMessageMode

is explicitly configured. Expected Behavior: The agent should generate and speak the first message using the configured custom voice according to the settings in

firstMessageMode

. Actual Behavior: No voice generation request is sent for the first message when a custom voice is set, resulting in a silent start to the conversation. Configuration Attempted: - Set

firstMessageMode

'assistant-speaks-first'

- Set

firstMessageMode

'assistant-speaks-first-with-model-generated-message'

- Custom voice configured as per documentation Documentation References: - Custom TTS setup: https://docs.vapi.ai/customization/custom-voices/custom-tts - First message configuration: https://docs.vapi.ai/api-reference/calls/create#request.body.assistant.firstMessage Test Call ID:

57ce58fb-11f2-4574-a2a2-aed65706db97

started at Jun 29 2025 10:48 UTC Additional Context: The issue appears to be specific to custom voice configurations, as the behavior may work correctly with default voices. The logs from the test call above should show the missing voice generation request for the first message.

Kfir Ochana

06/30/2025, 4:02 PM

@User @User Update: Working scenarios: - Direct MP3 file URLs for

firstMessage

(e.g.,

https://samplelib.com/lib/preview/mp3/sample-3s.mp3

) play successfully - Built-in OpenAI voices work correctly:

Copy code

json
"voice": {
    "provider": "openai",
    "voiceId": "nova", 
    "model": "gpt-4o-mini-tts"
}

Issue: Custom TTS models aren't playing the

firstMessage

content at all. Observation from call logs:

Copy code

[LOG] Voice input: [first message text]

Missing expected log:

Copy code

[CHECKPOINT] Assistant speech started

Actual problem: The TTS engine isn't being called for the first message (which isn't true for subsequent messages) when using custom voice configurations. Vapi appears to be skipping the first message entirely with custom TTS, rather than experiencing any latency or timeout issues.

Vapi

07/01/2025, 7:54 AM

Hi Kfir Ochana, To address the issue where the custom voice configuration is not triggering a voice generation request for the first message, ensure the following steps are correctly implemented: 1. **Voice Configuration**: Confirm that your custom voice is correctly set in the assistant configuration. Use the following structure:

Copy code

json
   {
     "voice": {
       "provider": "custom-voice",
       "server": {
         "url": "https://your-tts-api.com/synthesize",
         "secret": "your-secret-token-here",
         "timeoutSeconds": 30
       }
     }
   }

2. **First Message Configuration**: Verify that

firstMessage

and

firstMessageMode

are correctly set in the assistant's configuration:

Copy code

json
   {
     "firstMessage": "Welcome to our service.",
     "firstMessageMode": "assistant-speaks-first"
   }

Vapi

07/01/2025, 7:54 AM

1. **TTS Integration**: Make sure your TTS endpoint is configured to handle requests from VAPI and returns audio in the required PCM format. Here’s a snippet for the TTS server setup:

Copy code

javascript
   app.post('/api/synthesize', (req, res) => {
     // Handle synthesizing request
     res.setHeader('Content-Type', 'application/octet-stream');
     res.send(audioData);
   });

2. **Fallback Plan**: Consider adding a fallback voice to handle cases where the custom TTS might fail:

Copy code

json
   {
     "fallbackPlan": {
       "voices": [
         {
           "provider": "eleven-labs",
           "voiceId": "21m00Tcm4TlvDq8ikWAM"
         }
       ]
     }
   }

3. **Network Connectivity**: Ensure there’s no network issue between VAPI and your TTS server which might block the requests. By following these steps, the assistant should correctly generate and speak the first message using the configured custom voice. For more details, you can refer to the [Custom TTS integration guide](https://docs.vapi.ai/customization/custom-voices/custom-tts) and [Custom voices documentation](https://docs.vapi.ai/customization/custom-voices/custom-voice).

Kfir Ochana

07/01/2025, 12:14 PM

I appreciate you taking the time to help with this issue. However, I believe my situation goes beyond basic configuration problems. I've already implemented the voice and first message configurations correctly (as shown in my code snippet), and I've systematically tested several potential solutions. To clarify my current status: * The custom voice configuration is working perfectly** for all messages except the first one** * The first message configuration works correctly with built-in voices * I've already** tested the specific solutions you mentioned** (cachingEnabled=False, backoffPlan, and fallbackPlan) without success @User @User Since this appears to be a more complex integration issue rather than a configuration problem, could you please review the detailed observations I provided in earlier messages? I'm specifically looking for insights into __why the custom TTS server receives requests for all messages except the first one__, despite the configuration being identical. Any guidance on advanced troubleshooting steps or potential API-level issues would be greatly appreciated.

Kfir Ochana

07/01/2025, 12:14 PM

My code snippet:

Copy code

"voice": {
    "provider": "custom-voice",
    "server": {
        "url": TTS_SERVER_URL,
        "timeoutSeconds": 30,
        "secret": TTS_API_KEY,
        "backoffPlan": {
            "maxRetries": 3,
            "baseDelaySeconds": 2
        }
    },
    "cachingEnabled": False,
    "fallbackPlan": {
        "voices": [
            {
                "provider": "openai",
                "voiceId": "nova",
                "model": "gpt-4o-mini-tts"
            }
        ]
    }
},
# Set the first message
"firstMessage": "איזה כיף שענית לי!",
"firstMessageMode": "assistant-speaks-first"

Kfir Ochana

07/02/2025, 12:25 PM

@User Help, please!

Vapi

07/02/2025, 8:16 PM

Hey, Thanks for the details. Our team will look into it and let you know soon.

Kfir Ochana

07/03/2025, 12:54 PM

Thanks man! 🙌 @User Have you got any update?

Vapi

07/03/2025, 4:21 PM

Can you please get a latest call_id, for me to take a look into it.

Kfir Ochana

07/03/2025, 6:00 PM

20b31077-608f-4cd0-806d-7f39d235a1fd

Kfir Ochana

07/03/2025, 6:30 PM

@Shubham Bajaj

Vapi

07/06/2025, 7:38 PM

Hey Kfir, thanks for the details. Let me check with our team and get back to you soon.

Kfir Ochana

07/08/2025, 8:41 PM

Hi! any update?

Shubham Bajaj

07/09/2025, 8:54 PM

@Kfir Ochana I apologize for delayed, it slipped my list. I'm digging into it now.

Shubham Bajaj

07/09/2025, 9:11 PM

🟡 15:09:03:930 Couldn't Complete Say Request { "task": { "content": "Thanks for picking up the phone! Thanks for picking up the phone! Thanks for picking up the phone! Thanks for picking up the phone!", "endCallOnError": true, "interruptible": true }, "error": {} } 🟡 15:09:03:928 Couldn't Get Speech For

custom-voice:https://vu2xxxxxer.com/:16000:d763e003e95b009cad12e55e535fec922f15a56bd5b1da747a3cc8c5f4213b71

, "Thanks for picking up the phone! Thanks for picking up the phone! Thanks for picking up the phone! Thanks for picking up the phone!"

Shubham Bajaj

07/09/2025, 9:14 PM

@Kfir Ochana Let me look into your other call as well.

Kfir Ochana

07/09/2025, 9:17 PM

I'm available to try if you want me to replicate the scenario / try different things

Shubham Bajaj

07/09/2025, 9:26 PM

Can you confirm if this message is correct?

Kfir Ochana

07/09/2025, 9:27 PM

What do you mean by correct? My TTS server doesn't receive any request for generating firstMessage

Shubham Bajaj

07/09/2025, 9:27 PM

Was this expected message?

Kfir Ochana

07/09/2025, 9:28 PM

Yup : "Thanks for picking up the phone! Thanks for picking up the phone! Thanks for picking up the phone! Thanks for picking up the phone!"

Kfir Ochana

07/09/2025, 9:29 PM

It's phrased weird so I won't miss it (:

Shubham Bajaj

07/09/2025, 9:30 PM

Thanks for confirming. Let me look into the logs again.

Kfir Ochana

07/09/2025, 10:09 PM

@Shubham Bajaj Any insights?

Shubham Bajaj

07/10/2025, 1:31 AM

@Kfir Ochana Update: I couldn't find anything wrong, so I'm asking the team to take another look. A different point of view might help. What could happen going wrong here?

Shubham Bajaj

07/10/2025, 1:38 AM

@Kfir Ochana It's nothing to worry about. Sometimes, we can't immediately identify what might go wrong. So, it's better to seek another opinion rather than stubbornly sticking to one idea.

Kfir Ochana

07/10/2025, 1:40 AM

It gave me vibes of a tricky bug indeed

Kfir Ochana

07/10/2025, 1:41 AM

Appreciate how you keep trying 🙌

Vapi

07/11/2025, 5:51 AM

The request is taking a little longer than usual and we will continue to work on it. Thank you for your patience

Kfir Ochana

07/14/2025, 12:42 AM

@User any update?

Vapi

07/15/2025, 3:46 AM

Hey Kfir Ochana, Apologies for the delay. I have already raised your issue with the team, waiting for an update from their side. Once I received it, I'll let you know.

Previous Next