Latency Issue?
# support
d
Hi, my agent often isnt responsive for several seconds and needs to be reprompt by user to start speaking again. Here is one specific Call where it happened and Id like help to understand what is causing this specific issue. Call ID is 27a74a54-9e58-440f-a3ef-c377fe24a352 Issue happened at the 2nd conversation turn https://cdn.discordapp.com/attachments/1352292640573362236/1352292640791724093/Screenshot_2025-03-20_at_10.42.17_AM.png?ex=67dd7c29&is=67dc2aa9&hm=db26d3ab2b911148172e25c71a5b8feed15ad61be58efb3876d467a9e5d8a433&
u
Is this a phone call using twilio?
There have been (I'm having) issues with drift, where the caller and ai are talking over each other due to poor twilio audio delivery
d
This call is with Twilio yes, but I did not experience any drift or ai talking over
@User anyone from the team can look at this? These affect the quality of calls and happen quite often.
k
Hey @sandman, you need to update your assistant transcriber and point it to 300ms. Start speaking plan wait seconds should be set to 0.6. This will give your assistant an ample amount of time to identify if the user has stopped speaking, and then you can observe these unpleasant silences between the user and the assistant turn. Here's the API request: \`\`\`http PATCH /assistant/{assistant_id} Authorization: Bearer YOUR_API_TOKEN Content-Type: application/json { "transcriber": { "provider": "deepgram", "endpointing": 300 }, "startSpeakingPlan": { "waitSeconds": 0.6, "smartEndpointingEnabled": true } } \`\`\` This request will: 1\. Update the assistant's transcriber to use DeepGram with endpointing set to 300 milliseconds 2\. Set the start speaking plan's wait seconds to 0.6 3\. Enable smart endpointing You'll need to replace
{assistant_id}
with your actual assistant ID and
YOUR_API_TOKEN
with your valid API token. This request targets the PATCH endpoint for updating an assistant partially, which allows you to update only the specific fields you want to change while keeping the rest of the assistant configuration intact. If you need to specify any additional parameters or have any questions about this API request, please let me know!