ALS Stole My Father’s Voice. Could We Bring It Back?
“I just want to sound like me,” said Dad when he first started slurring.
Before he was diagnosed with ALS, Dad would listen to recordings of himself and declare that he didn’t like the sound of his own voice. Could have fooled me! Dad had always been a talker. On many occasions (often unasked) he had been a singer too. At the house I grew up in, his voice boomed through the walls at Christmas (“Come, they told him PA RUM PUM PUM PUM!!!”) At patient home visits for his work as a hospice doctor, his songs filled the silent spaces as he held their hands and sat with their families. At my wedding years ago, he went table to table singing “Hava Nagila” with aplomb. (He did not speak a lick of Hebrew.)
What Dad lacked in pitch and timbre, he made up for in volume and enthusiasm. Don’t know the melody? Just make it up. Don’t know the words? Nothing a little na na na can’t handle.
Once, he told me that when he was young he had been shy about being a poor singer, and had shrunk from any opportunity to sing. In middle age, a good friend who had long sensed Dad’s hidden passion encouraged him to come out of his shell. Dad was eternally grateful to this man and his portable karaoke machine.
Dad may not have liked the sound of his voice, but he had owned it and wanted to keep it.
I saw my mission clearly. After months of researching clinical trials, side-eyeing doctors and going down the “miracle cure” rabbit hole on social media, I had come to accept that I was not going to be The One Who Fixes Dad’s ALS. But! I had worked in tech for years. I grasped at the one thing I still had the hubris to believe I could be: The One Who Gives Dad Back His Voice.
Dad had read about message banks: platforms for patients to record their voice while they still can, speaking a set of predetermined key phrases like “I’m hungry” or “take me outside” or “I’m in pain.” When patients lose their speech and (most) movement, the message bank can be connected to an augmentative and alternative communication (AAC) device, designed for patients with limited mobility to trigger any one of those phrases when they want to.
HA! I thought, how primitive. My LinkedIn feed that year was frothing over all things artificial intelligence. Another day, another 10,000 mindblowing audio/visual demos:
We can make your video game character follow your eyes!
We can make the Mona Lisa say anything you type!
We can make anyone sound just like Drake!
In my mind, I could see all the pieces coming together: an eye-tracking tool to enable “typing” on a keyboard interface, text-to-speech conversion, an AI voice cloned from his recordings … and voila! He’d be saying all kinds of things in his own voice in no time.
To no one’s surprise except mine, this failed spectacularly, right from the start. We tried a few eye-tracking tools, only to find that Dad was too tired to focus his eyes to select letters on the screen. Then I brought home a stash of assistive communication switches (small sensors triggered by micromovements of a body part, like Stephen Hawking’s twitching cheek), but Dad couldn’t control any part of his body well enough anymore. I contacted shiny new brain-computer interface startups and research programs but no one had anything far along enough to try out.
Dad told me to stop trying. The thing about ALS, he said (slurred), is that even when you find something that just might help, it stops working eventually as you lose more and more control over your body, like the motorized wheelchair he used for a couple months, or the fingertip extender he typed with for a few weeks. Why pick up something new only to have it taken away after a short time?
The solution ended up being a human at his side (my mother, for hours and at all hours) with a notebook, scanning an alphabet chart printed on a laminated sheet while watching his face. Putting sentences together was slow and laborious, but this low-tech method worked better than my bounty of bleeding-edge toys.
But it was still not his voice. Ego deflated, I nearly abandoned the voice clone I’d started. I couldn’t use it to help Dad speak since we couldn’t get past that first step. Even if we had, like everything else, the clone just wasn’t turning out very well. Dad had no energy to make a proper training dataset, which required at least 30 minutes of clear speaking with clean audio. All we had were a few barely usable clips from noisy speeches at his annual hospice galas, voicemails he left for me over the years (including many takes of “Happy Birthday”), a decent rendition of “My Way,” and hastily recorded conversations I tried to catch before his slurring got bad. It was just enough to make a highly imperfect “instant clone.” Dad was a fast-talking Malaysian Chinese man; Instant Dad Clone sounded like a nervous South Asian man with traces of a British colonial education.
However! With no hopes pinning it down, Instant Dad Clone was fun. There was just enough of a glimmer of Dad to make “Instant Dad Clone says the darnedest things” an amusing activity for a couple of days. Instant Dad Clone extolled the virtues of Maggi mee, delivered the Gettysburg address, and told me he (it?) loved me.
Then I discovered voice-swapping. That summer, TikTok was flooded with Sabrina Carpenter and Kendrick Lamar wannabes. AI had made voice cloning for covers so easy and so illegal. I had smaller aspirations: an audience of one, whose tastes veered much further back in time. The first cover I made was one of his favorites: “Always On My Mind,” replacing the sultry tones of Elvis for the ambiguous accents of Instant Dad Clone.
Somehow, Instant Dad Clone’s accents, so jarring in speech, faded gently into the background in song. I hit play on my voice-swapped clip with no expectations and nearly fell out of my chair. It sounded like Dad. It was Dad. I could almost see him closing his eyes at the lilt of the verse (“Maybe I didn’t love you…”) and taking a deep breath right before the swell of the chorus (“you were always ON MY MIIIIIND!!”).
Dad was tickled. If anything, he sounded too good, he said. No way his real voice could handle those high notes.
We picked out more songs together. Dad had saved his favorite song tracks on his home computer over the years; to my mother’s chagrin when he was well he had enjoyed playing each one on repeat at full volume, and singing along (also at full volume). On his sickbed, he listened silently as we went through the stash to curate a playlist for Instant Dad Clone, and blinked when a song had his vote.
Songs by female vocalists needed to be pitched down to accommodate his register—otherwise Instant Dad Clone sounded like a scratchy, off-key Barry Gibb. Voice-swapping on live covers by non-professionals turned out better than studio-produced tracks by real artists; the lower production quality somehow suited Instant Dad Clone’s flaws. Not all experiments were successful. Instant Dad Clone singing Teresa Teng’s iconic “Yue Liang Dai Biao Wo De Xin” (“The Moon Represents My Heart”) was a disaster. But Instant Dad Clone singing “Devoted To You” by the Everly Brothers captured the wistfulness of the moment in Dad’s voice so perfectly that I couldn’t listen to it the whole way through.
In the end, Dad was right. With ALS, wins are short-lived. As he declined rapidly in his final months, Dad lost interest in voice-swapping and started to pick out songs for his funeral playlist instead.
A few months after Dad died, one of the voice cloning platforms announced that it was shutting down: “The landscape of AI-powered creative tools has evolved rapidly and we can no longer sustain the platform in a way that meets our standards and your expectations.” There were instructions on how to save our creations, if we wanted to keep them.
I did not. Instant Dad Clone’s 15 minutes were up. Dad himself had no desire for immortality. Once, I had mused that it would be easier to get an AI to learn all about him and speak for him rather than get his actual thoughts out of his head. Dad had not liked that idea at all—nor the implication that an AI could continue to speak for him after he was gone.
The only clips I saved were the ones I had uploaded to train the voice clone—still barely usable, but unmistakably Dad sounding like himself:
“For what is a man, what has he got?
If not himself, then he has naught
To say the things he truly feels
And not the words of one who kneels
The record shows I took the blows
And did it MY WAAAAAY!!”
Janine Yoong is the Chief Operating Officer at Remedy Robotics, a healthcare and physical AI company developing robotic systems for cardiovascular care and endovascular intervention. Her father, Dr. Michael Yoong, served for 15 years as chairman of Hospice Klang, a nonprofit in Malaysia that provides free palliative care to people with life-limiting illnesses.
All views expressed in this article are the author’s own.
Do you have a personal essay you want to share with Newsweek? Send your story to MyTurn@newsweek.com.
