Here is the full transcript of laryngologist Dr. Anthony Law’s talk titled “What Our Voice Reveals About Our Health” at TEDx Talks 2023 conference.
Listen to the audio version here:
TRANSCRIPT:
The Power of Voice
I believe that everyone in this room, all of you, are voice scientists. And as voice scientists, each one of you has the power to understand the health and the wellness of those around you. You see, I’ve dedicated my life to understanding how sound is correlated with health. I’m a surgeon and a scientist, and voice is a huge part of both of those practices.
You see, voice is our primary means of engaging with the world. It takes our innermost thoughts and turns them into concrete and shareable packages. But let’s be specific here. Voice is separate from speech.
Speech is the words that we choose and how we articulate it. While voice, voice is the result of the sounds that we make. Many of them turn into words, but some don’t, like ahs, ums, and sighs. Hopefully I don’t do any of that.
Because voice is a superset of speech, it contains more information than just the words that we choose. Henry here, one of my friend’s dogs, illustrates this point beautifully. You see, Henry doesn’t understand speech, but he does understand the intention and the emotion behind voice. “You bad dog!”
Understanding Vocal Cues
“You bad dog! What are you doing, you bad dog? Oh, you bad dog!” You see how Henry wags his tail and stares intently at the speaker, even though she’s saying pretty harsh things to him?
He’s responding to her voice and the emotion behind it. You see, we all unwittingly share a little bit of who we are and what we feel through the cues in our voice.
Through our inflections and vocal modulations, we can portray anger, fear, frustration, but also happiness, friendship, and love.
And as voice scientists, which we all are, we’re constantly analyzing the vocal cues in other people. We’re listening to whether a speaker is nervous or not. We’re making assumptions about how strong, how old, someone’s gender, and how confident they are. There’s actually new data out there that suggests a huge part of who we find physically attractive is influenced by their voice.
But arguably the most important thing we do when we analyze voice is we look for markers of wellness. Despite the fact that most of you in here haven’t studied in years in medical school, every one of you is pretty good at diagnosing sick versus healthy based off of a voice alone. My mom’s excellent at this. When I got COVID in 2021, within the first five seconds of our telephone call, she knew that something was wrong. She said she could hear it in my voice.
And we’re so very good at separating sick versus healthy because voice change due to illness is so very common. If we look at the numbers, one in three people in this room will have a voice change due to an illness in their life. It’s roughly that third of the room over there. And the reasons for these illnesses vary widely.
The most common reason is a common cold. But there are over 50 other individual diseases that result in voice change, ranging from seasonal allergies to Parkinson’s disease to cancer. My job is to go a step further than just separating the sick versus the healthy. My job is to explore the signals in voice that point to a particular disease.
And so myself and other people in my profession, we’ve become the tea leaf readers of voice. And we can do this because we have a fundamental understanding of how voice is made. We know that there are in general four structures that drive voice. The first, the lungs.
This is where the air and thus the power for voice comes from. That air is transmitted upwards towards the larynx or the voice box where the initial sound for voice comes from. That sound moves upwards again to the throat or the pharynx where it’s shaped and beautified. And ultimately, the teeth, tongue, and lips chop up the sound into intelligible packages.
Identifying Specific Diseases Through Voice
I know that particular aberrations at particular anatomic sites result in particular vocal cues. Thus, a dysfunction of the lung is going to sound markedly different than a dysfunction of the throat. And by studying the permutations of these vocal signals, I and others like me have been pretty good at pairing a voice change with a specific disease. Let’s listen to some voices.
“We were away a year ago. The blue spot is on the key again.” Do you hear that? How her voice is a little bit breathy.
And there’s some increased air flow. That’s because half of her voice box is paralyzed. Most likely she had a surgery and the nerve that controls her voice box was injured. Very typical for paralysis of half of your voice box.
Let’s try another one. “We were away a year ago. The blue spot is on the key again.” “How hard did he hit him?”
This voice change is a little bit more subtle. What I hear is that his voice is weak and that it lacks pitch variability. And ultimately I hear a little bit of that breathiness in his voice as well. Those cues come together to define the voice of Parkinson’s disease.
And that’s exactly what this patient had. And I want you to listen really closely to this voice. “Hello, doctor. How are you?”
“How are you doing today?” This gentleman came to my clinic a couple of months ago. I know from hearing his rough, strained, and gravely voice that there’s a high probability that he has laryngeal cancer. I also know, given the severity of these voice changes, that his cancer is likely in the advanced stages.
And if we talk to this gentleman and others like him, we’ll learn that his voice change has been present for months, sometimes up to a year. And likely the whole time that he’s had detectable cancer, he’s had voice changes. But unfortunately, neither he nor his primary care physician had the tools to identify cancer versus the other 50 diseases it could have been. If we look at the statistics for advanced stage cancer, like this gentleman has, they only have a 28% chance of survival at 5 years.
And to survive, they have to undergo chemotherapy, radiation, and more often than not, complete removal of their voice box. Fundamental change to the way they live. Where if we had caught the disease months earlier, in the very earliest stage of the disease, then treatment is a straightforward day surgery. And in early stage laryngeal cancer or throat cancer, nearly everyone survives if they’re treated appropriately.
The Need for Early Detection Tools
You see, the problem is not the absence of treatment for laryngeal cancer. The problem is we have an absence of tools for early detection. The overwhelming majority of voice changes are related to non-life-threatening diseases. The challenge becomes, how do we find the 1%?
The 1% that are from life-threatening processes and refer them urgently to a specialist for treatment. Right now, our inability to do this robustly results in 40% of the patients that come to my clinic already having advanced stage laryngeal cancer. That means that only 6 out of 10 patients with laryngeal cancer survive. Another stark statistic that’s particularly notable to me is the fact that most patients with laryngeal cancer, they look like me.
Black males have twice as likely to die from laryngeal cancer as their white counterparts. And as a lead of a lab that focuses on using software, computer science, and mathematical modeling to bridge healthcare inequities, I knew we had to get involved. So me and my team, we asked ourselves, what if the patient or the family doctor could perform a deep analysis on voice without the need for excessive specialty training? What if our family doctors had a tool that could identify high-risk voice changes and get those patients urgently referred to the right doctor at the right time?
Well, that’s what we’ve been working on for the last couple of years. Over the last couple of years, we’ve collected thousands of voices from a diverse set of patients, all ages, all ethnicities, some that are sick, some that are not, some with cancer, some without cancer. We’ve collected all these digital voice samples and we’ve built a deep learning, or AI, model. This model can discern who has a growth in their larynx or their throat and who doesn’t.
And right now, at present, our deep learning model can identify who has a laryngeal growth in their throat at about a 93% accuracy. We’ve taken our model from the depths of our basement and now we’ve deployed it onto a smartphone platform. That allows family medicine doctors to have easy access to our tool. And the best part is this.
The more voice samples we get, the more accurate our detection model becomes. So we’ve started to deploy our app in primary care clinics around Georgia. Our goal, we want to dramatically reduce the number of advanced-stage cancers that we have. And thus, we want to reduce the number of people that have to die from laryngeal cancer.
If we find benefit in Georgia, then we spread to the surrounding states, and then the nation, and hopefully the world. Our long-term goal is we want advanced-stage cancer to go the way of polio, a disease of a bygone time. So with deep learning, or AI, the involvement of thousands of patients in our study, and the dedicated effort of my group and others like us, we’ve slowly started to take the task of voice analysis and make it much more publicly available. We started by defining the acoustic cues of laryngeal cancer, but we’re continuing our exploration between voice and other diseases in multiple domains.
Each one of you can help us in our efforts to create accurate early detection by donating your voice. But more importantly, I want each and every one of you, the next time you’re at your family doctor or even home by yourself, if you notice a change in your voice, I want you to step back and think, what does my voice say about me? Thank you.