Improved Training for End-to-End Streaming Automatic Speech Recognition Model with Punctuation