I recall reading somewhere that Stephen Hawking's original voice was a DECtalk, specifically a voice included with it called "Paul". I'm guessing that, in order to keep this voice, someone had to dump the samples from a DECtalk or even call up HP to see if they could buy the masters. The article says he copyrighted the voice, so I'm willing to bet it was more the latter.
It's interesting to note that until 2001, the "Paul" voice was used by NOAA to broadcast weather alerts to automatic weather radios. It's also interesting to note that the voice itself was a series of samples recorded by Dennis Klatt, a researcher who dedicated his career to helping the disabled speak again and passed away shortly after the DECtalk was released. That Hawking was able to adopt it as his own, is a rather sweet tale with a cyberpunk twist.
the master from his voice never belonged to HP, it is owned by Nuance after multiple acquisition. The original owner was Speech Plus, who made the CallText 5010 synth he used until this year.
They indeed were looking for the masters [1].
When they found it they made a replacement for the voice synth on software running in a raspberry pi. They had just replaced the the old voice synth back in Jan. [2]
Additionally, Dennis Klatt was the inventor of DECTalk, the three original voices were his own for the male voice, his wife voice for the female voice and his daughter voice for a child voice, they were named Perfect Paul, that was the default voice, Beautiful Betty and Kit the Kid.
> the three original voices were his own for the male voice, his wife voice for the female voice and his daughter voice for a child voice, they were named Perfect Paul, that was the default voice, Beautiful Betty and Kit the Kid.
Given Ramachandran's Phantom Limb experiments, the "hit a fake hand with a hammer" experiment, and an experiment done by a crazy/brilliant Swede[1], I think it's very plausible that Hawking though of the synthesized voice as his voice. (I mean, he intellectually knew that it wasn't, but that doesn't make the feeling go away.)
[1] I can't find a link for the life of me, but I think I saw it on BBC's Horizon programme where a person wears a camera helmet which is showing a different person's perspective which includes the wearer as a "third person". In a few minutes the wearer starts to feel dissociated from their own body and starts to feel associated with the "viewer". It's a really weird and interesting experiment. Hopefully, this'll spark a memory and someone can find a video.
Thank you, but that's just the "plain" phantom limb illusion. I was thinking of a completely different documentary -- the person was actually wearing a complete VR setup. (Now that I think about it, I think it might have been a researcher at Uppsala University. Just in case anyone finds this comment in future.)
Ah fantastic - I hadn't seen that one ... - it's also not available at present in the UK from that BBC link, but it gave enough info for me to find it at https://www.youtube.com/watch?v=X09TgwcfQyA -- hopefully that works for you, thanks.
This actually makes a lot of sense since the Speak and Spell toy we all grew up with has a similar voice to Hawking. The link you provided shows that they used related synthesis models
A little OT, but I think squarely in HN territory:
Years ago, a friend of mine's mother-in-law lost her voice due to cancer. As an engineer, he built her a tool from a toy Casio keyboard and some samples, allowing her to talk using this portable toy.
After that experience, he talked about wanting to create a service by with which people could archive their voices, so that in case of future need they could be loaded as samples for such a device. He'd give them scripts to read and record, from which he'd parse the phoneme samples, and then storing them in a repository that they could later retrieve if necessary.
Today there's probably a need for a lot more security than he'd talked about at the time, but it seems like something that's eminently do-able today. It could be a very cheap and low-cost endeavor, but useful as a sort of insurance because it carries a really high value in the unusual cases where it's needed.
Does anyone know about using one of these to create an offline archive of a voice? Are these things always connected to the internet? I really love the idea of saving voices of loved ones as if they are photographs. Or of moments in time (your child's voice, etc.)
I experimented with Lyrebird the other night, recording the maximum 300 sentences for their learning algorithm, and I wasn’t impressed with the output. With that said, big fan of the idea and I hope they can improve the output over time.
I didn't actually have a great sense for how fast a normal person might talk. In case anyone else is curious, it turns out that normal conversation averages about 110-150 words per minute. That definitely puts into perspective how sluggish 10 wpm is.
Someone clarify because I'm confused. I thought there were newer alternatives with better hardware but Hawkings wanted to keep his old system. How does giving way his system help when there are better implementations out there?
Hawkins wanted to keep the familiar voice he has been using. He did not want it replaced with 'better' voices because he came to identify strongly with the voice.
They (Intel?) replaced the hardware and software with a better system, called Assistive Context-Aware Toolkit (ACAT). It allowed him to use computer easily (compared to the old system) [0]:
The new Intel user interface operates like an engine. A sensor mounted on Hawking's cheek is detected by an infrared switch attached to his glasses, helping Hawking select a character on his computer. A customized version of the popular SwiftKey app has greatly improved the system's ability to predict his next characters, words and actions so he only has to type the first few characters of a word. That information is relayed to his speech synthesizer.
To search the Web, for example, Hawking previously had to take lengthy routes: Exit from a communication window, navigate a mouse to run the browser, navigate the mouse again to the search bar, and finally type the search text. The new system automates all of these steps seamlessly and quickly.
He refused new voice synths because the voice in the newer implementations were different enough that he would not consider his voice.
They eventually found the master from his voice and developed a new synth in software that run in a raspberry pi that he was using since jan 2018.
the only thing they kept was the voice synth, everything else was replaced. What was developed by intel and made open source was the software he use to interact with the computer named acat [1]
It's interesting to note that until 2001, the "Paul" voice was used by NOAA to broadcast weather alerts to automatic weather radios. It's also interesting to note that the voice itself was a series of samples recorded by Dennis Klatt, a researcher who dedicated his career to helping the disabled speak again and passed away shortly after the DECtalk was released. That Hawking was able to adopt it as his own, is a rather sweet tale with a cyberpunk twist.
You can compare the DECtalk "Paul" voice with Klatt's own voice here: http://www.festvox.org/history/klatt.html