Breaking the sound barrier

摘要

Classic offline subtitling of recorded programmes uses a conventional keyboard to retype the dialogue subtitle-by-subtitle, timing it as you go. This is laborious and time-consuming. Methods borrowed from live subtitling now mean that real-time transcription techniques (such as re-speaking to trained speech recognition engines) are being used to capture the text - and indeed to capture the timing as well - all in one pass. If a script is available, text-to-speech alignment tools can be used to synchronise the script to the audio in order to generate first-pass timecoded subtitles.