Duration-aware pause insertion using pre-trained language model for multi-speaker text-to-speech