Need for automatically generated narration

Abstract

This paper argues that the best current text-to-speech (TTS) synthesis systems are approaching the quality necessary to provide effective automated narration of audio books. Currently, nearly all audio books and audio journals are recorded by professional voice actors at great expense with significant lead times. These cost and time constraints mean that fewer than than 2% of the new titles published each year are available in "Talking Book" editions, leaving the visually-impaired and print-disabled community of users with few options when seeking material in digital libraries. State-of-the-art TTS systems now can reproduce human voice prosody of sufficient quality to make listening to long narrative reading both pleasant and comprehensible. Such technology is relatively compact and inexpensive; it is time to deploy it widely as an alternative means of accessing digital texts. This would not only directly benefit the reading-disabled community, but also enable "digital natives" and other users to listen to texts on platforms on which reading may not be practical.