Customer Reviews

Text to Speech Synthesis: New Paradigms and Advances (Prentice Hall Imsc Press Multimedia Series)

5 star:		(0)
4 star:		(1)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

The field of TTS has been steadily improving. But still not perfect. If you listen to an extended TTS audio, you are unlikely to imagine it was a single human recording. Here, the editors provide a set of research papers that map out the boundary of TTS.

What I found the most interesting was the chapter comparing it with Automatic Speech Recognition. The latter is a much harder problem. Especially if you want speaker independence. And the input audio can have noise. Whereas TTS is effectively noise-free. The input text is always precisely known. But the chapter points out an ironic difference that is somewhat of a mirror image. ASR accuracy can be easily and objectively measured, by comparing the ASR's output text with the text transcribed by a human listener. Whereas the "goodness" of a TTS audio output is very subjectively determined.

This is one major unsolved TTS problem.

The chapter goes into some of the ASR methods that have been brought successfully into TTS research. Most notably is the use of Hidden Markov Methods. In ASR work, this was perhaps the biggest innovation in the last 10 years. It also shows some preliminary promise for TTS.