[Final Call Transcript] Why AI is not precise?
# support
I'm using Vapi for an Italian solution, and the AI transcript AI is not 100% accurate to the spoken audio by the AI. Why don't use the results of the LLM API calls and save those? Looks like that you re-transcript the whole audio again, and this will create more errors in the transcripts.
We don't know when the bot will be interrupted so use transcribed bot audio to get ground truth. Using https://developers.deepgram.com/docs/keywords should resolve inaccuracies related to names, etc.
Hi, I have the same issue, and it's not a STT issue. Transcript is really different from logs, it becomes generally plain wrong in french. I posted here: https://vapi.canny.io/feature-requests/p/inconsistencies-between-logs-and-transcript Everything is working well otherwise, but this issue is preventing me from using Vapi.
@Godefroy @Lonestone Ask this question on Vapi Office Hour https://discord.gg/KnK58ue6?event=1240493862867304479
@Sahil Thanks but I won't be available at this time, and I think it's a real bug you should address
First message of my assistant : "Bonjour, je suis l'IA de Lonestone" In the transcript : "Un jour, j'assuse Léa d'un lone stone" It's like you're doing STT to the voice generated by TTS, it's absurd.
@Godefroy @Lonestone First of all, sorry for the delay. Coming to your question, can you create a new thread and provide me with the call_id + timestamp of the voice recording and the expected output and all that.