Australian start-up Laronix is developing a bionic device that uses AI-based voice-cloning technology to restore the voices of laryngectomy patients.
When Dr Farzaneh Ahmadi was studying for her PhD in Singapore a decade ago, she received an email that has stayed with her ever since.
Working as part of a team researching bionic voice technology, she was contacted by a laryngeal cancer patient in the US.
“A patient emailed our group and said, ‘I’m facing this condition; I will lose my voice and I’m a singer — my voice is my life. Can you help me?’”
Ahmadi was distressed by the fact that the research wasn’t advanced enough to restore the patient’s original voice.
“Our honest answer was, maybe in 20 years,” she says.
“At that point, it started to bother me personally. Why can’t we solve this — something that seems so biologically possible?”
Today, Ahmadi spearheads Queensland-based start-up Laronix, which is developing an artificial intelligence (AI) supported bionic voice that promises to not only enable speech for people who have had their voice box removed but also give them back their own, distinctive voices.
Globally, 184,615 people were diagnosed in 2020 with laryngeal cancer, the most common head and neck cancer. Some of these patients will undergo a laryngectomy — voice box removal — and become known as laryngectomees or, more affectionately, “larys”.
Although lifesaving, the surgery can have a big psychological impact. As Ahmadi puts it, “society is impatient with voice disability” and, sadly, depression and suicide are common among survivors.
Laronix’s Bionic Voice, a smart device that combines non-invasive hardware and artificially intelligent voice-cloning technology — the same as used for deep fakes — aims to offer hope to people who have exhausted other options.
Old-school solutions
During a laryngectomy, the voice box is removed, and the trachea — the windpipe — is moved to the front of the neck, where a stoma is created to allow breathing.
As anatomy differs between patients, voice rehabilitation after surgery must as well.
“There is huge variation [among patients],” says Alanna Bowen, an independent speech pathologist who works with Laronix.
“It’s individualised and it’s purely down to how people heal from surgery, the amount of treatment they need to get, their own capabilities, dexterity, support people. It’s a patient population that has to be treated individually because there’s just no one size fits all.”
Existing solutions include voice prostheses and the electrolarynx, a device that is activated when held to the neck and which sounds quite robotic. While many laryngectomees have success with these, others run into problems, says Bowen.
Tracheoesophageal voice prosthesis, considered the current gold standard, features a one-way silicone valve inserted into the tissue between a patient’s windpipe and oesophagus. When a patient covers their stoma, this device redirects air to the back of the throat, which vibrates and creates sound.
Though it can produce a strong voice, it must be cleaned and changed regularly, posing issues for some patients.
“When I was working clinically for public health, I would have patients travel 500 km to come and see me for a voice prosthesis change,” says Bowen.
“I was the closest person who had the skill to be able to give the service that they needed.”
While she was studying, her own grandfather underwent a laryngectomy. After first trying a prosthesis, he then used an electrolarynx.
“[My grandparents] had a business in town and it took a long time for people [to adjust]. Even though they knew my grandfather, they knew he’d had this surgery, and it was a little country town — if he would pick the phone up for the business, people would hang up.”
Bowen says, while there have been tweaks to these solutions, the technology hasn’t advanced much in decades.
Ahmadi says one reason for this is the location and complex structure of the larynx.
An unexpected find
Originally from Tehran, where she’d worked as a telecom engineer, Ahmadi came to Australia via Singapore in 2013.
Frustrated with the lack of progress, she moved away from bionic voice research.
When applying for a fellowship to instead research optimality in biology, she got a knockback that proved to be an unexpected source of inspiration.
“One of the reviewers had said, ‘If biology is so optimal, why aren’t you using that principle in generating voice for people, which has not worked in your PhD?’ That was, ironically, a eureka moment,” Ahmadi recalls.
From there, she moved to the University of Sydney. When undertaking an extensive literature review, she stumbled across research in China that described a pneumatic artificial larynx (PAL) — a very basic mechanical larynx that uses a person’s respiration signals to create voice.
“It’s the world’s simplest model of human larynx — it’s out of the body, completely mechanical,” says Ahmadi.
“Ironically, it generates Chinese language, which is a tonal language, meaning that it can generate pitch variations inside the word, something that healthy people also find difficult to do.”
This was a crucial find.
“It was the ultimate proof in our hand — that the only thing that you need to convey an exceptionally high-quality voice for humans is respiration; you don’t need any commands from the brain. That immediately simplified the problem,” Ahmadi says.
By 2018, Ahmadi and a team at the University of Western Sydney had developed what they called a Pneumatic Bionic Voice, modelled on the PAL, which underwent pre-clinical trials.
Ahmadi and her brother, Dr Mousa Ahmadi, founded Laronix in 2019, as a spin-out from the University of Western Sydney.
Last year, the start-up won a major category in Bionics Queensland’s Bionics Challenge, which aims to accelerate bionics development in Australia.
Director of Bionics Queensland Dr Robyn Stokes says she was impressed with Laronix’s robust research and development.
“Where it stands out is the quality,” she says. “The clarity and the non-invasive aspect of it and possibly how far [they’ve] gotten in terms of tapping what that artificial intelligence can do in that context — it’s much more advanced [than other solutions].”
There may also be less obvious applications for the technology, says Stokes.
“It has uses beyond for those with laryngectomy. [For example,] there are millions of dollars lost economically from the loss of voice in the education sector.”
AI Assistance
Unlike surgical solutions, Bionic Voice is a non-invasive electronic device that uses a sensor on the neck to monitor signals generated when a user breathes.
This information is sent to an AI module enclosed in a headset, which acts like vocal cords.
In a recent model, the voice is then transferred to the mouth via a thin tube — but Laronix aims to further streamline the design. The ultimate goal is to give people their natural-sounding voices back, says Ahmadi.
“Once you generate voice, the second step is to train that voice to become more natural. That’s where the AI kicks in,” she says.
Using voice-cloning, the AI algorithm is fed audio clips and can be taught to mimic a particular voice in five to 20 minutes, improving in quality over time.
Users will be able to alter the pitch of their voice as well as sing and laugh — and one day, thanks to the AI, it may even sound like they did pre-surgery.
Acceleration
Ahmadi says taking a product from a university to the market hasn’t been a totally smooth ride. But the company has attracted significant public and private investment and trials are underway this year.
Stokes says, under Ahmadi’s leadership, the company is well placed for global acceleration: “She completely understands the global marketplace”.
Ahmadi remains driven to help as many people as possible.
“For every one of our patients that gets their voice back, you’re just ultimately so thrilled,” she says. “At some point, it becomes a matter of numbers – how many lives you can save.”
Finding a voice
One patient who has been overjoyed to try the Bionic Voice is laryngeal cancer survivor, Bobbi (last name withheld for privacy).
She was previously using a voice prosthesis, but because she’d sustained tissue damage in her neck from radiation, the valve became loose and began to leak, causing serious complications.
Before working with Laronix, Bobbi had been silent for several years, which poses daily frustrations and has taken a huge toll on her mental health.
“I use [an e-writer] or a text-to-speech app on my phone, but both forms of communication are so much slower than verbal communication that by the time I have written my thoughts the conversation has moved on and what I have written is no longer relevant,” she says via email.
Using the Bionic Voice, Bobbi says she could “be part of society again”.
“Socialisation would be easily possible; parties and meetings and groups would be accessible. My mood would lift enormously,” she says.
“I could put the device on and order coffee verbally, I could go to the butcher and ask for items, I could tell my daughter she looks lovely today, I could yell at my grandson, I could train my dog, I could go to dinner with friends and participate in the conversation. I could be/feel normal again.”