Yes this is a known gap with current voicemail detection. Carrier voicemail that reads the phone number aloud often gets transcribed as normal speech, so the assistant responds before detection triggers. I can help fix this by adding a pre-response guard that looks for early numeric read-outs + voicemail phrasing before the assistant speaks (not relying on LLM speed). This avoids longer pauses and random hangups. Quick question: 👉 Do you control the first audio frames / transcription stream (before the assistant responds), or are you reacting only after the full transcript comes back?