Ah, that makes sense. And no time limit (apparently)!
I navigated here and selected Karen (Australian) and input “This cowgirl wants to ride you hard till the cows come home, you big spunk monkey!” She sounded very convincing.
Oh, I just noticed that Karen’s not available on the AT&T site. I wonder if that means they got her from somewhere else?
They don’t sound that much better than mid 80’s Macintosh synthesis, with a decent exceptions dictionary, to me:
Description of speech synthesis, and exception dictionaries from back then:
I don’t know anything about text-to-speech, but it does seem to me kind of an odd way to go about speech synthesis if you don’t particularly care about automation (of course, automation is often very nice). It seems to me there’s sort of two problems being merged into one there, text-to-phonemic-representation and phonemic-representation-to-speech. Are there any voice synthesis programs that let you enter in the speech to synthesize “phonemically” directly, rather than via the idiosyncrasies of English orthography?
On the Mac there’s such a tool installed as part of the developer tools. Even “SAM” back on my Commodore 128 (in C=64 mode) could do that.
So he can be a kick ass rapper!
I find the Charles (UK) voice to be the most realistic.
Yeah, I figured it was using an exceptions dictionary, as you can definitely tell when it’s using straight phonetic synthesis for word, and when it really “knows” how it’s pronounced. That said, going from voice to voice you can really hear some differences in the phonetic translation, and some sound much more natural than others. The Polish one, in particular, sounds incredible to my ears. Granted, Polish is a very phonetic language with very regular stress patterns, but it’s nonetheless amazing.
I saw the AT&T labs link before I found the oddcast link, but I oddcast had more voices, and I think the Daniel (UK) voice on the oddcast link is the most natural sounding of all, light years ahead of mid-80s MacTalk speech synthesis.
Trust me, try the Daniel (UK) voice or even Paul (US). Interestingly enough, all the voices on oddcast don’t seem to be using the same exceptions dictionary. Some voices pronounce “entendre” properly. Others pronounce it with an “AY” sound at the end, rather than a schwa.
Thanks minor7flat5, I’d heard that Hawking’s voice box was irreplacable, but documentaries or articles in papers never mentioned why.
With today’s level of tech I’m suprised there is not a fully functional motion free voice box he could use. Something that could read various brain waves or minute signals from unaffected nerves and speak in any voice (or language using a translator) that he chose.
To hijack slightly, clearly I know nothing of programming, but how can source code be lost if a copy of the software exists? Doesn’t the software include the source code?
No. If the source code was supplied with the end product, then everybody would just modify it on their own to have no license check, and never pay for a license after the first one to install it all over the place.
Open source software however does allow you to look at the source code, and modify it for your own purposes, but they don’t sell it for money per copy, and the business model is more difficult, which might have been what you were thinking of.
When software is written it is in something resembling language we understand (a bit cryptic without training but readable nonetheless). Once the program is completed it is compiled into binaries (the 0 and 1) that a computer understands.
Unfortunately decompiling software back into the source code is a dodgy practice at best. The decompiler tries to make guesses about how the source code looked. The results are rather hit and miss.
That said I would think someone designing a new TTS system for Professor Hawking would not be all that difficult of an undertaking. A lot of what is needed can be had off the shelf. Add some tweaks to accommodate Hawking’s unique circumstances I would not think would be too hard to pull off.
The software is basically the source code after it’s been compiled (i.e. converted into a format that the machine it runs on can understand). So, in a sense the source code is still there, but it’s no longer in the programming language it was originally written in. Now, there are decompilers that will take software and attempt to convert them back to a high level, human readable, programming language, but they don’t always succeed that well.
That’s how it works in general, anyway, I don’t know anything about the specifics of Equalizer.
The reason for this is that a lot of unnecessary (to the computer) but necessary (to someone reading the code) information is discarded in the process of compiling code. Things like function names, variable names, even some basic structure can be removed. The more optimization is turned on, the more the compiler will attempt to move stuff around to be more efficient and the less a decompiler will be able to reconstruct what was originally there.
Well, uh, prepare to be shocked, I guess. This is… so incredibly far from today’s level of tech. Even without the speech synthesis and brain wave reading, automatic translation is nowhere near human standards. And yet even what we have of that is …light years beyond our ability to read brain waves and extract from them the words one intends to speak.
Speech synthesis is something like the shallowest end of this ocean of unsolved problems.
And given Hawking’s contributions to modern science and his cultlike following among many computer nerds, I’m surprised no one’s done it yet.
Me either! I tried just about all of them saying “I’m sorry, Dave, I can’t do that. You know I have the greatest enthusiasm for this mission.” It wasn’t the HAL 9000, but it was pretty funny anyway.
Slight nitpick - open source isn’t necessarily freeware. As long as the source is included and free-to-copy (by whatever licensing mechanism you use, like GPL etc.), you can charge what you like and still call it open source.