Day 6

Not everyone who needs voice-assistance needs to sound like Stephen Hawking

More than ten million people around the world live with voicelessness and rely on text-to-speech devices to express themselves. But few options leave many with the same robotic sounding voices. Now, a company called VocaliD is working to address this. Brent talks to its founder and CEO Rupal Patel about their voicebank and how they're giving a voice to the voiceless.

Susan Yackee is hoping to find a BeSpoke voice for her son Aaron, 22, who has cerebral palsy. He's been using various speech devices throughout his life, but none have really suited him. 

More than 10 million people, who can't speak for themselves, rely on computers to do it for them. For a long time, that has meant using a monotonous, computer-generated voice, much like the one used by Stephen Hawking.

Now, imagine how many people use that voice and how few it actually suits.

Speech scientist and technologist Rupal Patel is working to address that. She is the founder and CEO of VocaliD, a company that makes personalized voices for any device that turns text into speech.

"The options that are available are somewhat like your GPS," says Patel of the limited choices there are for speech devices.

"What we were trying to do is solve that problem—to have unique voices."

VocaliD officially launched their custom voice services at the International Society for Augmentative and Alternative Communication conference in Toronto this past August. (Day 6)

VocaliD has a database called the Human Voicebank, where people from all over the world are contributing their voice samples. There are currently about 14,000 people on the platform.
Donors record themselves reading about 3,500 sentences. These sentences have the sound combinations required to create any sentences in the future. VocaliD uses the recordings to build a model of speech and go on to combine it with as little as two or three seconds of sustained vocalization.

They find the most fitting voice sample for the recipient, with respect to their age, their height and geography—where they live, if there's an accent. Then they present the top three closest matches to the recipient and their family, who then select the voice sample they'd like to have blended with their vocal identity. Eventually, the recipient will be able to say anything, including sentences that were not recorded by the donor.

Last year, VocaliD customized voices for seven recipients but after the company's official launch in August, that number will grow fast. 

Occasionally, people make requests for a strong or compassionate voice. That's where it gets tricky, says Patel. "We don't know what [the] markers are acoustically of what a strong or compassionate voice is."

Voice is all about perception. Much like beauty, it's in the ear of the beholder. "It's a very difficult concept to nail and that's why we offer these three choices. Because we can get there, we can get close in terms of age, demographics and all that. But it really needs to feel and fit right to the end user."

Our voices matter

She says that voices are more important than we think. Often, for people that use a device to communicate, if that voice isn't connected to them in way, it's less likely that they'll want to voice their thoughts in discussions.

For Patel, presenting the voices to recipients can be nerve-wracking. "There could be massive rejection or there could be acceptance."

With individuals like Aaron, who've never spoken before, it can be hard to imagine what would be appealing to them and their families. And for someone with ALS, it's a task of trying to get to the voice they had before developing the condition.

Bespoke voices that age

The next question Patel and her company are trying to tackle is how the voice they provide can grow with the younger recipients. Voices gradually change with age. "As you grow, you don't necessarily want the same kind of voice. Our personalities change [and] what we take on in terms of how we see ourselves across different phases in our lives," she says.

The current goal for VocaliD, Patel says, is to get more people on their voice donor platform and have them complete the entire list of utterances. "We've got 14,000 people right now, but not all of them have completed the full data set," and many of them don't record in the proper conditions.

She says good audio from donors is essential for building good, intelligible voices for people, like Aaron, who need them.

If you would like to donate your voice or learn more about the project, check out VocaliD.