Play.ht voices keep glitching and stop talking and the Agent stops responding Vapi AI #support

Play.ht voices keep glitching and stop talking and...

kitten

02/24/2025, 5:31 PM

Seems the system stops recognising any input because I am talking but it doesnt show in the recording either. Call ID: 37266b81-b59d-4260-84ab-aa542d8afdb9 Timestamp: 5 sec in Its not hearing me talk.

kitten

02/24/2025, 5:48 PM

Heres an other one where it actually heard me but wont reply: 694cfc55-c6ef-439f-bddc-6c3acbeebd88

kitten

02/24/2025, 5:58 PM

This also happens on calls where it repeats the word and keeps saying the same word for a very long time. a0be6d62-4921-44e9-b40e-bbdee05dcb62 at 30s

kitten

02/24/2025, 10:44 PM

@User

kitten

02/24/2025, 10:44 PM

I need this fixed please!

Kyle

02/26/2025, 9:53 AM

Kitten, your assistant is not not recognizing input and hears but won't reply s in required time because of endpointing The default endpointing and startSpeakingPlan config.

current_config

- transcriber {"model": "nova-3", "language": "en", "provider": "deepgram"} - start_speaking_plan1 {"smartEndpointingEnabled": true}

next_config

Copy code

{
  "provider": "deepgram",
  "model": "nova-3",
  "language": "en",
  "endpointing": 300  // Increased from default to prevent premature cuts
}

Copy code

{
  "smartEndpointingEnabled": true,
  "waitSeconds": 0.8,  // Increased to give more processing time
}

Kyle

02/26/2025, 9:53 AM

- The default deepgram endpointing value of 10ms is too aggressive for many use cases. From the Deepgram documentation in the code:

Copy code

/**
 * - Defaults to 10. This is recommended for most use cases to optimize for latency.
 * - 10 can cause some missing transcriptions because of the shorter context. 
 * - This mostly happens for one-word utterances. For those uses cases, it's recommended to try 300.
 */

Kyle

02/26/2025, 9:54 AM

These changes give the system more time to: - Process the complete utterance - Detect proper sentence endings - Avoid premature interruptions - Ensure the full context is captured before responding

Kyle

02/26/2025, 9:54 AM

here's how you can update the transcriber and start speaking plan settings: curl --location --request PATCH '' \\ \--header 'Authorization: Bearer YOUR_API_KEY' \\ \--header 'Content-Type: application/json' \\ \--data-raw '{ "transcriber": { "provider": "deepgram", "model": "nova-3", "language": "en", "endpointing": 300 }, "startSpeakingPlan": { "smartEndpointingEnabled": true, "waitSeconds": 0.8, } }'

kitten

02/26/2025, 1:53 PM

What about this issue? also happens on calls where it repeats the word and keeps saying the same word for a very long time. a0be6d62-4921-44e9-b40e-bbdee05dcb62 at 30s

Kyle

03/02/2025, 2:38 PM

Hey @kitten sorry for the delay, can you check the call ID again I don't see any repeating words?

Copy code

"transcript": "AI: Our calling demo line is down for maintenance at the moment. Please check back later.\nUser: Okay.\nAI: A moment while I do a few checks. Break Let's do it. First things first, can you share your name and email with me?\nUser: Sure. It's brandon and brandon 1 2 3 at g mail dot com.\nAI: Some Brandon. So\n",

Previous Next