Synthetic voices can be a terrifying object these days when combined with deepfakes and other AI deceptions, but they are also an essential tool for those who can no longer speak for themselves. Acapela Group is looking squarely at these people with its new “My Voice” service that allows anyone to train his AI voice profile for free.
Acapela has been in the text-to-speech space for almost 25 years, was recently acquired by technology accessibility giant Tobii Dynavox, but still operates independently.
As with many industries, accessibility has been greatly impacted by the emergence of consumer-scale machine learning processes. Seven or eight years ago, Acapela’s co-founder Remy Cadic recalled that customizing synthesized speech for himself was not only cumbersome, but the results weren’t great.
“It was very time consuming. Patients had to train for eight hours. Now I can bank my voice with just 50 recorded sentences. will be ready,” he said. “There is definitely a revolution going on in neural text-to-speech technology.”
Having a voice generator that uses your own voice is certainly something that more and more people can appreciate. Choosing from a list is a little impersonal. Many people have a voice they’d rather use, but it’s only recently that it’s become an option.
They weren’t kidding about how quick and easy it was. It was 50 short sentences, and articles. The recording interface is simple and easy to navigate, and indeed, after a day or so, my voice was ready to use. It’s clearly your voice and can handle the sentences you throw at it on the demo page.
If you now need it, you can pay to download it and use it with any compatible speech generation system. Obviously, this includes Tobii Dynavox’s TD Talk and devices. In fact, the company just released a new one last week. These are pretty sophisticated.
Delaina tests the new Tobii-Dynavox device.
And that’s the real point of it all. This is not a technical demonstration of the power of neural voice technology, nor a demo where anyone can feed in the voice of a celebrity to clone. A tool specially created for people who have only had complex processes.
We know that many people facing a degenerative disease, cancer, or certain procedures may not be able to speak well or at all within months or years. Making the process of banking your voice as easy as possible is a service many people appreciate.
“One of the big advantages is that it can also be customized for children. , and we’re still heading in that direction,” said Cadic.
Recording and re-recording or artificially aging banked voices is a new and challenging feature, but it seems to yield results:
Compatibility with offline devices that don’t have the latest neural processing chips is also a key differentiator. “There are online solutions for creating voices easily, but they are only available via the cloud and are not practical,” he said.
By the way, 50 sentences is great for someone who can still read and write, but you can also train your voice with voice recordings of people who have lost that ability. It’s not that simple.
The company also realized that diversity and thoughtfulness in the training process are as important as any other AI application. The problem with the ultrafast training technique, Cadic points out, is that it “just tries to find the speaker in the training material that is closest to the user. I can’t hear you.”
Acapela Product Manager Nicolas Mazars added that like many AI problems rooted in poor training data, this problem is not evenly distributed. Are you African American or do you speak English poorly? We work in 23 languages and have many disabled users. We rely on user feedback and try to develop something for them. ”
The recording and banking process is free. Sign up for an account here and train your own synthetic voice in minutes. You only pay when you download and install it on your device.
