Recently we faced with serious degradation using nova-2-general transcriber models. We've changed nothing from our side but still some stuff seems to be broken using the same model. These are some examples: 1. Call ID: 2df19c34-82e8-4ea5-8ae1-76ce66edc370 Transcriber hears

Hi

messages multiple times while user didn't say anything like this. (example at 1:10) 2. Call ID: d869bf08-5b8a-43f0-97a3-bb934cfddb26 Transcriber didn't hear

Welcome Okay

message (0:14) 3. Call ID: d869bf08-5b8a-43f0-97a3-bb934cfddb26 Transcriber hears

You wanna pull back

but user didn't say this (02:14) 4. Call ID: d869bf08-5b8a-43f0-97a3-bb934cfddb26 Transcriber didn't hear full sentence

Okay. My previous reservation uh, was also supposed to be on Friday, But at 2 PM

(3:22) 5. Call ID: 8fefca93-bbad-458d-b179-20a96820c76f Transcriber heard

I want to confirm my reservation.

while user said

I want to cancel my reservation

(0:13) 6. Call ID: 2df19c34-82e8-4ea5-8ae1-76ce66edc370 Transcriber heard

Thank you. Ida. Please speak your, uh, SMS.

while user said

I just received an SMS

(1:37) 7. Call ID: f9a680b1-4766-40cd-b229-abb14866d676 # that was actually nova-2-phonecall Transcriber heard

I’m a real dick tuna

while user said

I am allergic to nuts

(4:08) Just for clarification - we haven't seen as much failed transcription before, we're thinking of some sort of degradation coming from transcriber backends/VAPI side, please provide us any help with that!

@anyhero I'm sorry to hear about the issues you've been experiencing with the transcription accuracy. It's important to ensure that the service you rely on is performing optimally. To address the concerns you've raised, we'll need to investigate the specific instances you've mentioned. For each of the call IDs provided, we can review the transcriptions and audio to understand what might be causing these inaccuracies. It's possible that there could be a temporary issue with the model or an underlying problem that needs to be addressed by our technical team. Since you've noticed a pattern of degradation, it would be helpful to gather more details about the audio environment, any changes in the audio quality, or if there have been any updates on the transcriber's end that could have affected the performance. In the meantime, I can forward these examples to our technical team for a detailed analysis. They will look into the transcriptions and the model's performance to determine the cause of these issues and work on a resolution. If you have any additional information or further examples, please feel free to share them as they could be helpful in our investigation. Rest assured, we are here to help and will do our best to resolve this matter promptly. Please ask support for more info on this.