Custom TTS Help
# supportm
muertos
08/08/2025, 4:12 PMHey team! I have an agent configured to override with a custom TTS server in certain cases in my app. The voice override wll look like this:
{
provider: "custom-voice",
server: {
url: ttsUrl,
secret: secret,
timeoutSeconds: 30,
},
};
For some reason i'm not seeing any requests come through to my server when I initialize the SDK with this override.
const assistantOverrides = {
// Voice configuration
voice: voiceConfig,
// Custom LLM configuration
model: {
model: "gpt-4o",
provider: "custom-llm",
url: process.env.VAPI_LLM_URL,
},
// Metadata for tracking
metadata: {
agentId,
sessionId,
isPublic: true,
},
};
What could I be doing wrong? Thanks for all your help!
David
v
Vapi Support Bot
08/08/2025, 4:13 PMThanks for sharing your config! To help debug, could you confirm how you are starting the call or triggering the TTS in your app (e.g., which SDK method or function you’re calling after setting
assistantOverridesm
muertos
08/08/2025, 4:29 PMHow we start the call and trigger TTS:
1. SDK Initialization: We use the Vapi Web SDK (@Vapi-ai/web) with our public
key:
const vapi = new Vapi(process.env.NEXT_PUBLIC_VAPI_PUBLIC_KEY);
2. Call Flow:
- User clicks to start a call in the UI
- This triggers startAudio() in our VapiProvider component
- We first fetch the assistant configuration from our backend at
/api/vapi/start
- The backend returns the assistantId and assistantOverrides (including
voice config)
- We then start the call using:
await vapi.start(assistantId, assistantOverrides);
3. Assistant Overrides Structure (from /api/vapi/start/route.ts):
const assistantOverrides = {
voice: {
provider: "custom-voice",
server: {
url: "https://our-backend.com/tts/stream/{agentId}",
secret: process.env.BUYER_INTELLIGENCE_TTS_SECRET,
timeoutSeconds: 30
}
},
model: {
model: "gpt-4o",
provider: "custom-llm",
url: process.env.NEXT_PUBLIC_VAPI_LLM_URL
},
metadata: {
agentId: "...",
sessionId: "...",
isPublic: true
}
};
await vapiRef.current.start(config.assistantId, config.assistantOverrides);
5. Event Listeners: We have standard Vapi event listeners set up:
- call-start
- call-end
- speech-start
- speech-end
- error
- message
Assistant ID: d1bf4b7c-b2cf-45c3-aff6-f8593d62b924 @User
muertos
08/08/2025, 4:34 PM@Vapi Support Bot
v
Vapi Support Bot
08/08/2025, 4:35 PMThanks for the detailed info! To help narrow this down:
Could you check if there are any errors or failed network requests in your browser’s console or network tab when you start the call with the custom TTS config?
This will help determine if the SDK is attempting to reach your TTS server or if the issue is earlier in the flow.
Sources:
- [Vapi Web SDK Documentation](https://docs.vapi.ai/sdk/web)
- [Custom Voices Documentation](https://docs.vapi.ai/customization/custom-voices/custom-voice)
m
muertos
08/08/2025, 5:10 PMThere are no errors unfortunately @Vapi Support Bot The sdk initializes fine and "Thinking" state begins
v
Vapi Support Bot
08/08/2025, 5:12 PMSuggestion:
Double-check that your assistant configuration includes a
firstMessage Copy code
js
const assistantOverrides = {
voice: { /* your custom-voice config */ },
model: { /* your model config */ },
metadata: { /* your metadata */ },
firstMessage: "Hello! This is a test of the custom TTS."
};firstMessagem
muertos
08/08/2025, 8:13 PMstill nothing... after connecting the sdk just stalls and I dont get any requests to my backend
server: {
url: 'https://[redacted]/tts/stream?user_id=329da444-5e0e-440c-bfcf-ab5ee0d8c423&voice_id=cj8CnlzTyo5TBdEyr9Lq',
secret: redacted,
timeoutSeconds: 30
}
muertos
08/08/2025, 8:13 PM@Vapi Support Bot
v
Vapi Support Bot
08/08/2025, 8:14 PMTry this minimal override to rule out config issues:
Copy code
js
const assistantOverrides = {
voice: {
provider: "custom-voice",
server: {
url: "https://your-backend.com/tts/stream",
secret: "your_secret",
timeoutSeconds: 30
}
},
firstMessage: "Testing custom TTS connection."
};m
muertos
08/08/2025, 8:22 PMhere is some more code: try {
// Get assistant configuration from server (includes voice settings)
const config = await startCall(assistantId);
if (!config) {
throw new Error("Failed to get assistant configuration");
}
console.log("[SecureVapiProvider] Starting Vapi with exact config:", {
assistantId: config.assistantId,
assistantOverrides: JSON.stringify(config.assistantOverrides, null, 2),
});
// Start the call with the assistantId and overrides
// The overrides include voice settings but no API keys
// API keys cannot be passed from frontend for security
await vapiRef.current.start(
config.assistantId,
config.assistantOverrides,
);
} catch (err) {
setLastError(err as Error);
setConnectionState("disconnected");
} @Vapi Support Bot still nothing when I use: await vapiRef.current.start(config.assistantId, {
voice: {
provider: "custom-voice",
server: {
url: "https://backend-url/tts/stream",
timeoutSeconds: 30,
},
},
firstMessage: "Testing custom TTS connection.",
});
v
Vapi Support Bot
08/08/2025, 8:24 PMNext Diagnostic Step:
To confirm your TTS server is publicly accessible, try sending a test request from a public network (not your local machine) using `curl`:
Copy code
bash
curl -v "https://backend-url/tts/stream"m
muertos
08/08/2025, 9:50 PMOk so I was able to get a response. But it took so long, well over 30 seconds. Most of the time it hots the Vapi disconnect timeout. Why isnt vapi calling my TTS server quickly?
muertos
08/08/2025, 9:50 PM@Vapi Support Bot
o
ooo1993
08/10/2025, 11:39 PMCustom TTS is available only for pro or enterprise plans?
m
muertos
08/10/2025, 11:53 PMis this true?
c
Chiranjeet Mishra
08/11/2025, 10:47 PMThis is not true. CustomTTS is available for all plans as long as it is configured properly. To check why the TTS took a long time to respond, please check the logs and you can provide them here too if you would like us to review it
o
ooo1993
08/12/2025, 2:57 PMIv'e tried to set custom-tts to my agent using API like that following the docs:
Copy code
curl -X PATCH https://api.vapi.ai/assistant/ASSITANT_ID \
-H "Authorization: Bearer TOKEN" \
-H "Content-Type: application/json" \
-d '{
"name": "New Assistant",
"voice": {
"provider": "custom-voice",
"server": {
"url": "URL",
"secret": "SECRET",
"timeoutSeconds": 45,
"headers": {
"Content-Type": "application/json",
"X-API-Version": "v1"
}
},
"fallbackPlan": {
"voices": [
{
"provider": "11labs",
"voiceId": "5kmbtrsekikrzsdxxrzg"
}
]
}
}
}'c
Chiranjeet Mishra
08/12/2025, 10:44 PMThis happens due to
custom-tts