Transcript messed up
# support
f
Hi i just tested a call where the assistant spoke first something, then i did and then again assistant did and then again i did speak something. It was clearly distingushable in real life call, but when i look at the transcript its messed up and overlapping. I see multiple such occurances. Please help with this. Call id: 9340f55e-dffa-4158-befb-96c1e8780525
v
Hi there! 👋 I understand how frustrating it can be when transcripts don't accurately reflect the conversation flow. I can't access the specific transcript for your call ID, but I can definitely help explain what might be causing the overlapping issues and suggest some troubleshooting steps. ## Potential Causes for Overlapping Transcripts There are several reasons why your transcript might be showing overlapping speech when the conversation was clearly turn-based: 1. **Endpointing Issues**: The system might be prematurely detecting the end of your speech, causing transcript segments to appear out of order. According to [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1337135526981140621), this is a common cause of overlapping transcripts. 2. **Audio Channel Mixing**: Sometimes the audio channels for the assistant and user get mixed, making it appear as if both parties are speaking simultaneously in the transcript, even when they weren't. This was mentioned in [Discord support threads](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1316042862630146058). 3. **Transcription Model Issues**: The transcriber model might be struggling with speech patterns or audio quality, leading to misaligned transcripts. Some models like Deepgram's nova-2-phonecall have been reported to have specific issues with transcript accuracy according to [user reports](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334475576152428546). ## Troubleshooting Steps Here are some steps you can take to resolve the issue: 1. **Adjust Endpointing Settings**: - Increase the duration the assistant waits for you to finish speaking - Configure custom endpointing rules using REGEX as suggested in [this thread](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1337135526981140621) - Try increasing the
waitSeconds
parameter to 0.8 in your configuration
2. **Try a Different Transcriber**: - Some users have resolved similar issues by switching transcribers - You could try alternatives like Whisper or Azure if you're currently using Deepgram - Update your transcriber settings with a command like:
Copy code
curl --location --request PATCH 'https://api.vapi.ai/assistant/YOUR_ASSISTANT_ID' \
   --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,
     }
   }'
3. **Check Audio Quality**: - Ensure there's minimal background noise - Check that your microphone is properly configured Your call ID (9340f55e-dffa-4158-befb-96c1e8780525) can be used by the Vapi support team to investigate the specific issue in more detail if these steps don't resolve the problem. Would you like me to explain any of these solutions in more detail or help with implementing any specific fix? Sources: - [Discord discussion on endpointing issues](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1337135526981140621) - [Discord thread on audio channel mixing](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1316042862630146058) - [Discord thread on transcriber model issues](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334475576152428546) - [Discord thread on troubleshooting steps](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1343636482874933340)
p
Hey! To help track down this issue, could you share: - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help us figure out what went wrong!