Summary: CereVoice and Cepstral are too crash-prone, laggy and inaccurate to be bothering with for any serious uses. Stick to what Apple provides, plus Acapela as a fair second.
I had sufficient data on my 3G plan that I thought I’d get the two TTS engines. The trials were inadequate to test screen reader performance, of course. And the test strings always sound much better than anything you provide, if for no other reason than that reading is a very different experience from using a screen reader. The costs are not too favourable. There are 19 English-speaking CereVoice voices, at £25.99 each (about $39.70). Then there’s a per-voice care package, at £17.99 each (about $27.48). For Cepstral, there are 21 voices including a few novelty voices such as dog and Whispery, each going for $35. For comparison, InfoVox iVox voice credits can be bought, one a voice for installation on one computer (yay for the curse that is activation!), for $14.99 each, and there are discounts for purchases of 5 or more (20%), 15 or more (30%), 30 or more (40%) and 125 or more (50%); if I were to purchase the 40 voices I now have for one Mac, it would cost just $359.76. You download the voices from the web pages at the sites; Cepstral voices are also demos, while CereVoice requires you log into an account and use a short-lived but refreshable download link (10 minutes) for each voice. VoiceOver couldn’t activate the links inside tables, as it so often doesn’t; I had to save the page source and tamper with it to make the links clickable. As you may imagine, the latter presented some difficulty for me on 3G; I ended up scripting the retrieval of the downloads on a remote machine and then downloading them to my local machine at my own leisure. Both engines are provided as OS X installer packages, one per voice, but nothing like iVox; there are no central accounts, only a license key for each voice. Cepstral was markedly better here, with the installers being at least scriptable using the installer command, and the keys entered in System Preferences; CereVoice popped up a license dialog during installation that requested a fiddly multi-line key that you copied and pasted from email, so while it could be mostly automated, one had to be careful to dump information into Terminal to find out which package you were licensing. The packages contain the voice, the engine and the license management framework, so duplication and wasted download time and bandwidth. Still, there is something to be said for avoiding having an agent watch over your shoulder using the Internet activation system, as is the case with iVox. Tragically, Cepstral made the supreme mistake of calling their TTS engine “swift”, and the even more unfortunate misstep of installing a binary, /usr/bin/swift, to offer command-line playback. And yes, that overwrites the binary put their by the operating system and Xcode. This OS installation is trashed. As to the voices themselves, it was very refreshing to see so much British diversity in CereVoice, but mostly because the company is based in Scotland and therefore presumably liable to get more British and Scottish voices. Cepstral, by contrast, only really offered two UK voices, and the rest are US english. Obviously this is a concern more for off-worlders (i.e., people not living in the US) but I’m not really familiar enough with the accents of other English-speaking countries outside the US or UK, so I can’t really say. All of the voices are passable for different kinds of reading, CereVoice more so than Cepstral, in my opinion, but they are simply unusable for VoiceOver. They lag, freeze, crash, overlap themselves in times of stress—in short, only use for dedicated reading applications. My favourites are CereVoice Sarah and Cepstral Laurence. Maybe I will reinstall these voices in future, for my entertainment or if they are updated. CereVoice additionally charges for support and maintenance, which I ultimately paid. However, I don’t recommend anybody else get them. The costs aren’t really justified for that. Some might argue that any money spent on TTS is too much, or that we don’t have real choices or anything formant, but at least iVox gives you a nice, responsive alternative to the incumbents. Perhaps Mac OS isn’t a platform either vendor would prefer to support; certain things, like the way the packages are installed, certainly suggest it. Still, as I said at the top, I don’t think it’s worth the bother, and there’s a very strong likelihood that the day I reinstall these voices will never come. -- You received this message because you are subscribed to the Google Groups "MacVisionaries" group. To unsubscribe from this group and stop receiving emails from it, send an email to macvisionaries+unsubscr...@googlegroups.com. To post to this group, send email to macvisionaries@googlegroups.com. Visit this group at http://groups.google.com/group/macvisionaries. For more options, visit https://groups.google.com/d/optout.
