Hey @Abdullah Yes, this is a common issue with Arabic TTS/agents. The model usually treats long numbers as a single token, which causes fast or unnatural pronunciation. The fix is to force number chunking (digit-by-digit or grouped) before speech output—either via preprocessing (e.g., 7 2 3 5 6 1 2 3 or 72 35 61 23) or by explicitly converting numbers to spoken Arabic text in logic, not just the prompt. I’ve handled similar voice agent setups and can help you implement a clean, natural-sounding solution depending on the TTS or platform you’re using.