Hi guys, Thanks for reading this. I'm having trouble with getting consistent voice output. It often hallucinates (= says gibberish, no real words). I don't understand why, since it's a high quality voice and my settings are conservative. Here's my code: "voice": { "provider": "11labs", "voiceId": "1h6gHcRvQugDNZgdAo9N", "language": "de", "stability": 0.9, "model": "eleven_turbo_v2_5", "similarityBoost": 0.4, "style": 0 } Are there any other settings to make the voice more reliable? Appreciate it a lot. Tom

@tom Hello Tom, I'm glad you reached out for assistance. Your current settings seem well-configured for a high-quality voice output. However, if you're experiencing issues with the voice hallucinating, there are a few additional settings you might want to consider: 1. **Input Minimum Characters**: Ensure that the

inputMinCharacters

setting is appropriately configured. This setting determines the minimum number of characters that will be passed to the voice provider, which can affect the consistency of the voice output. The default is 30 characters, but you might want to experiment with this number to see if it improves the output. 2. **Input Punctuation Boundaries**: The

inputPunctuationBoundaries

setting can also play a significant role. By specifying punctuation marks, you can help the voice provider understand sentence boundaries better, which can lead to more coherent speech output. 3. **Audio Format**: Check the audio format and bitrate settings to ensure they are compatible with your voice provider's recommendations for the best quality. 4. **Contextual Information**: Providing more context to the voice model can sometimes help with the clarity and relevance of the output. 5. **Provider-Specific Settings**: Some voice providers have additional settings or flags that can be used to fine-tune the voice output. Check the documentation for your specific provider to see if there are any other recommended settings. If you'd like, I can help you adjust the

inputMinCharacters

and

inputPunctuationBoundaries

settings. Would you like to proceed with that, or do you have specific values in mind for these settings?

Here is the complete assistant code (removed prompt and ids): { "assistant": { "name": "Outbound Demo Agent", "voice": { "provider": "11labs", "voiceId": "1h6gHcRvQugDNZgdAo9N", "language": "de", "stability": 0.9, "model": "eleven_turbo_v2_5", "similarityBoost": 0.4, "style": 0 }, "model": { "model": "gpt-4o", "toolIds": [ "", "" ], "messages": [ { "role": "system", "content": "" } ], "provider": "openai", "temperature": 0.8, "knowledgeBase": { "topK": 2, "fileIds": [ "" ], "provider": "canonical" }, "emotionRecognitionEnabled": true }, "recordingEnabled": false, "firstMessage": "Hallo, spreche ich mit {{2.

1

}} {{2.

2

}}?", "endCallFunctionEnabled": false, "transcriber": { "model": "nova-2", "language": "de", "provider": "deepgram" }, "clientMessages": [ "hang", "function-call", "speech-update", "metadata", "conversation-update", "tool-calls", "tool-calls-result" ], "serverMessages": [ "end-of-call-report", "hang", "function-call", "tool-calls" ], "serverUrl": "", "maxDurationSeconds": 555, "backchannelingEnabled": false, "backgroundDenoisingEnabled": false, "messagePlan": { "idleMessages": [ "Hallo?", "Sind Sie noch dran?" ], "idleTimeoutSeconds": 7.5 }, "startSpeakingPlan": { "transcriptionEndpointingPlan": { "onNoPunctuationSeconds": 0.5 } }, "stopSpeakingPlan": { "backoffSeconds": 1 } }, "phoneNumberId": "", "customer": { "number": "" } }

@tom did you end up solving this? if so, how? were having the same issue but its intermittent