@AiTitus :) The model may hallucinate if the transcript is empty. Alternatively you can also try handle this in your prompt by specifying a default message for such cases. However, we will consider your feedback and add it to the feature request list.