Although all the voices on their site have the same version number, there are qualitative differences between them -- the two new voices, David & Diane, are the best in my opinion. Another, Duncan, is also very easy to understand.

Overall, this product is much better in quality and value than the Apple voices (although they are "free"), or even Speechissimo , which I was using before (Enan cut down the quality of the Elan voices used in Speechissimo.) However, I wish some of the other voices they offered were of better quality. For example, the British female voice, Millie, is a bit choppy and 'pops' frequently when speaking, when I did some real-world tests on my machine.

I also think the default speed is too fast for a couple of voices -- Millie sometimes skips over words such as "the" -- fortunately, the speech rate is adjustable on a voice by voice basis in Apple's speech preference pane. I had to take Millie down half a notch to make her more understandable.

How do these compare to the holy grail of TTS voices, AT&T voices? Well, AT&T still holds the prize for quality -- the AT&T British female voice, Audrey, runs circles around Millie, for instance. But OS X availability of the AT&T voices is currently limited to a programmer's SDK starting at $300. Quality of the Cepstral voices is still the best I've heard, second to the AT&T voices. In terms of cost and system integration, Cepstral wins by a mile.

I highly recommend these voices. I'm currently using David to proofread my writing and to read long news and science articles in the background. Cepstral also offers demo downloads of each of it's voices, which is a good way for you to become familiar with them before deciding which to buy.

Finally, I think these voices raise the bar enough to show the short comings of enabling/controling speech from within OS X, and I hope to see some apps in the future that make controlling TTS a bit easier.