Apropos of my last post, here’s an sample of the same paragraph, part of an earlier post here, rendered in various text-to-speech systems. All samples have been converted to MP3 files (using iTunes) for ease of comparison:
- IBM Text-to-Speech: Unconstrained U.S. English Text Demo
- AT&T Natural Voices Demo
- Festival Voice Demo
- Winbond USA TTS Demo
- Testen Sie die ATIP Sprachsynthese interaktiv
- Mac OS X Built-in Speech
To my ears, the IBM female voice is the best sounds — it has a certain melodic quality that’s quite pleasant. The AT&T speakers are very good too, if a little over-zealous. Winbond can’t really be faulted for its poor quality — it’s a “TTS on a chip” system, not a general purpose software system.
Comments
The “Festival Demo” voice is the one which sounds the most like what I imagined a computerized voice would sound like — like Stephen Hawkin’s.
It also reminds me so much of the “Fitter Happier” track from Radiohead’s OK Computer record — eerie, actually. Most of the other samples sound quite natural, all considering.