Hi all! I wrote this prompt to capture the user’s email address, but the problem is that the transcriber sometimes doesn’t write down some of the letters I spell — I can tell because they don’t appear in the log. Also, when the email address is a bit longer or includes difficult names, even after multiple repetitions the agent still can’t understand the email correctly. Another issue is that when the user takes a bit longer to spell the email, the agent starts speaking in the middle of the user’s spelling, which interrupts the process and makes it hard to finish properly. This ability to capture the user’s email address accurately is really important to me. By the way, these are my current settings — I’d be happy to hear any ideas you have for improving them: - Model: GPT-4o mini cluster - Temperature: 0.4 - Voice: Eleven Labs Turbo v2-5 - Stability: 0.7 - Clarity + Similarity: 0.75 - Speed: 1 - Style Exaggeration: 0.4 - Transcriber: Deepgram Nova 3 DE - Confidence Threshold: 0.3 - Start Speaking Plan Wait seconds: 0.3 - Stop Speaking Plan Number of words: 2 - Voice seconds: 0.2 - Back off seconds: 1