Conventional automatic speech recognition systems do not produce punctuation
marks which are important for the readability of the speech recognition
results. They are also needed for subsequent natural language processing tasks
such as machine translation. There have been a lot of work