At the forefront of Artificial Intelligence

Home » Reviews » Software » Speech Technologies

AT&T NaturalVoices (Desktop Edition)

AT&T NaturalVoices Desktop is simply a collection of text-to-speech voices of an incredibly high-quality. The package is completely SAPI compatible, so any SAPI enabled program (for instance, HomeSeer) will be able to use them.

I received the package on no less than seven CDs, one for the main installation program that installs two low quality voices (8Khz), and six others that contained other voices, or duplicate voices of higher quality (16Khz). It quickly became apparent that the installations for the voices are very big - about 500Mbs per voice. The final installation program (with 4 voices available in 8Khz and 16Khz format) stands at 2.81 Gigabytes. With HDD space being so cheap these days, it isn't all that much of a problem, but for HDD-conscious people, or people working with a space quota, it might cause problems.

Quality

Thankfully, the quality of the voices more than makes up for the size requirements. The voices all have very distinct accents, intonation differences, and pitch ranges. As the saying goes, the proof is in the pudding - just listen to these clips:

Microsoft Sam (for comparison) - WAV, 89Kb
Mike @ 16Khz - WAV, 160Kb
Rich @ 16Khz - WAV, 157Kb
Claire @ 16Khz - WAV, 204Kb
Crystal @ 16Khz - WAV, 170Kb.

It is still evident that the voices are computer generated, but compared to Microsoft Sam, they definitely are more realistic. I tried using the voices to read website text, long passages and e-mails. Using the Microsoft voices, it quickly became tedious or hard to follow. The AT&T voices make listening much easier, probably due to their more natural speech patterns.

Mike

Mike is definitely my favourite, since his voice sounds the most realistic. His intonation and speech patterns tend to be slightly more realistic than the others. He has (for want of a better phrase) the "friendly American" style of talking, making it easy to listen to.

Crystal

Crystal is also good, but she tends to sound a little shy. As stupid as I feel writing that, its true! Her voice doesn't flow as well either, making her sound a little choppy.

Rich

Rich has a great voice, and definitely shows how AT&T's technology can produce a very large range of voices. He has a definite style and accent, but I felt it sounded a little weird coming from a computer!

Claire

Claire reminded me of an old woman. The voice is very good, but sounds strained and wobbly and times (thus the old woman). Perhaps a teaching program would benefit from it, but I'd prefer my "You have mail" message to sound soothing, not commanding!

The quality of all four voices is equally impressive, and choosing a voice will be a matter of taste. AT&T also provides users with the option of creating your own voices (for an additional price) and the technology also supports additionally languages. The NaturalVoices website (see below) has examples in Spanish, German and more.

HomeSeer

I received the package from the kind people at HomeSeer to use with their superb home-automation program. Since the voices are completely SAPI compatible, it is simply a case of selecting the voice and you're away! The voices sound great in HomeSeer, since you normally require them to say complete sentences (such as the demos above), but not huge passages that might show the weaknesses of TTS.

The only strange behaviour I saw was that Mike16 and Claire16 were swapped around. For example, if I selected "Mike16", Claire's voice would speak the "Hi, I sound like this. Testing 1, 2, 3" and vice versa. That aside, things worked a treat.

Developers

The Desktop edition even comes complete with all the necessary library and header files to create your own applications using their SDK. The demonstration applications are a little confusing and aren't Windows based (all console), but some console demonstrations are better than no demonstrations at all!

Conclusion

Principally, the two things that might put people off are the price and the size. The pricing scheme is a little complicated - you must purchase the Mike/Crystal set first ($56.95 from HomeSeer), then other voices are $19.95. Note though that this is much cheaper than buying through AT&T ($99 for Mike/Crystal, $49 for voices).

Again, the size for all four voices was well over the 2 Gigabyte mark. Our test machine has 120Gb available to it, so it isn't too bad.

For this price and size requirement though, the AT&T voices are brilliant. If you don't mind your computer sounding like one, then don't bother shelling out the cash. If you enjoy HomeSeer and TTS technologies, NaturalVoices is for you. My recommendation would be to stick to the basic package first since Mike and Crystal are the best voices currently available.

9.1

Price: $99 for base voices. $49 for others.

Liked: Excellent TTS quality. Large range of voices.

Disliked: Price and installed size.

Website: http://www.homeseer.com/

Submitted: 04/06/2002

Article Toolbar

BibTeX entry