In real-time voice generation, the system generally cannot set the temperature too low, otherwise the voice will become monotonous and mechanical.