Your Voice Matters: Using Speech to Monitor Huntington’s Disease

Disease progression in HD is currently monitored primarily through tests scored by clinicians, neuroimaging, and measurements made in blood or spinal fluid. But a group of researchers from France have recently developed an automated device that can estimate how HD is progressing, by analyzing a person’s speech while they perform a simple counting task. 

Remarkably, measures of speech production could accurately predict the results of clinical assessments in people with HD and successfully track changes over time. These findings demonstrate that voice analysis might represent a promising, noninvasive, and real-world tool for tracking disease progression. Tools like these could someday make it possible for people with HD to monitor their condition using a smartphone, reducing the need for frequent trips to the clinic, lowering healthcare costs and helping doctors provide more personalized care.

Everyone’s voice is unique

Our voice is an individual trait – no two people sound exactly alike. Some voices are highly recognizable. Readers hailing from an era before cell phones and caller ID will recall picking up a land line and identifying a friend’s voice on the other end. Younger readers might recognize their favorite actor behind a cartoon character, or Taylor Swift’s dulcet tones on the radio. Many factors make each individual’s voice unique. These include the shape of the vocal cords, the size of the throat, mouth, tongue and sinus cavities and how sound bounces around an individual’s head and neck. How you speak is also affected by your age, health, local accent and personal style of talking.

Why can humans talk, but not animals?

Glad you asked. The human voice, as we know it, is like a wind instrument, relying on three main things: sound, resonance and articulation. Sound comes from the vocal chords. Your vocal cords are two flexible, V-shaped bands of muscle tissue and mucous membranes inside the larynx, better known as the “voice box”, which is located in the middle of your neck. During regular breathing, your vocal cords stay completely open so air can pass into your lungs. But when you speak, your brain signals the voice box muscles to pull the cords shut, causing them to vibrate and thereby creating the sound waves of your voice. These sounds are modified by resonators, the throat, mouth cavity, and nasal passages, which produce a person’s recognizable voice. Finally, the tongue, soft palate, and lips are considered to be articulators that modify the voiced sound and are important for articulating vowels versus consonants. 

Mammals, reptiles, and even amphibians have similar vocal cord anatomy as humans. These animals communicate with one another using noises and sounds, such as barks, roars or screeches, but only humans can speak words. This is due to direct neural control of the voice box, tongue and lips and specialized cortical areas in the brain that allow for conscious control over our sounds, rather than instinctual noises. Researchers study these neuronal connections to understand how the brain controls speech and language, and how it can go wrong in disease states.

Speech complications in HD

Speaking involves complex motor abilities, cognitive control, and planning at multiple levels, so it is probably not surprising that speech difficulties have been documented in people with symptomatic HD for decades. Early studies showed that individuals with HD had slower rates of speech, took longer to say words and produced greater silences between and within words, compared to healthy controls. Specific changes also include increased pitch or harshness of the voice and reduced coordination of the tongue and lips, making it more difficult to articulate words. Interestingly, some reports have also suggested that speech deficits may precede the onset of the first overt movement symptoms. Hence, it is thought that tracking changes in speech may be a valuable marker of disease onset and progression. 

Your voice as a biomarkerbiomarker a test of any kind – including blood tests, thinking tests and brain scans – that can measure or predict the progression of a disease like HD. Biomarkers may make clinical trials of new drugs quicker and more reliable.

A “biomarkerbiomarker a test of any kind – including blood tests, thinking tests and brain scans – that can measure or predict the progression of a disease like HD. Biomarkers may make clinical trials of new drugs quicker and more reliable.”, or biological marker, is formally defined as a characteristic that is objectively quantified and evaluated as an indicator of a normal biological process, disease state or a response to an intervention. Simply put, it’s something you can tangibly measure that gives information about what’s happening inside the body due to disease or in response to a treatment. Biomarkers play a crucial role in multiple aspects of HD care and drug development. Importantly, they can be used to assess disease onset, monitor disease progression or determine how well a new therapy might be working. Currently, the most common biomarkers for HD include proteins that can be measured in cerebrospinal fluidCSF A clear fluid produced by the brain, which surrounds and supports the brain and spinal cord. (i.e. neurofilament lightNfL biomarker of brain health [NfLNfL biomarker of brain health] and huntingtin proteinhuntingtin protein The protein produced by the HD gene. [HTTHTT one abbreviation for the gene that causes Huntington’s disease. The same gene is also called HD and IT-15]), or brain imaging biomarkers (such as magnetic resonancemagnetic resonance A technique using powerful magnetic fields to produce detailed images of the brain in living humans and animals imaging [MRImagnetic resonance A technique using powerful magnetic fields to produce detailed images of the brain in living humans and animals]) that determine brain volume. 

But instead of invasive spinal taps or expensive neuroimaging, wouldn’t it be great to have an easy, noninvasive way to track disease symptoms or how a clinical trialclinical trial Very carefully planned experiments designed to answer specific questions about how a drug affects human beings medication is working, on a regular basis and potentially remotely? Recent studies suggest that this may be possible! 

Predicting clinical scores with a short speech test

A standard visit to a neurologist for someone with HD will involve a wide range of assessments of motor, cognitive and functional performance.  From these clinical scores, clinicians have  developed the composite Unified Huntington’s Disease Rating Scale (cUHDRS), which combines features of motor, cognitive and functional symptoms into a single unit. The cUHDRS is a widely accepted way to track the progression of HD and is commonly used as an endpointendpoint A specific outcome or measurement that researchers use to assess the effectiveness or safety of a treatment. Endpoints are predefined before the trial begins and can be either primary (the main result the trial is designed to evaluate, such as improvement in symptoms) or secondary (additional outcomes of interest, such as quality of life or biomarker changes). in HD clinical trials. But as with any test to track HD, there are drawbacks: this assessment is often only done once a year, it isn’t designed to capture change very sensitively, and the results can differ based on who is conducting it. HD, after all, is tremendously complex. 

In cutting-edge studies published in 2022, a team led by Anne‑Catherine Bachoud‑Lévi recruited 181 HD gene carriers from three French prospective cohorts to provide simple speech recordings. These recordings involved counting numbers 1-20 forward and then backwards in less than 45 seconds. Counting backwards is actually much harder and involves more working memory, focus and attention, so this was an important addition to the study. The results showed that certain speech features could predict the main clinical symptoms measured by the cUHDRS. Further, MRImagnetic resonance A technique using powerful magnetic fields to produce detailed images of the brain in living humans and animals imaging data showed that deteriorating speech seemed to track with damage to the striatum, the part of the brain that is most affected by HD. 

What the new study showed

Building on this previous study, Bachoud‑Lévi’s team recently developed a tool called xHD-Vox (x for eXplainable, HD for Huntington disease, and Vox for voice) to track and analyze voice recordings. They first automated the speech recordings using “Whisper”, an automatic speech recognition system – yup, artificial intelligence (AI). Whisper was created by OpenAI, the same company that developed ChatGPT.

Like any AI-based tool, xHD-Vox had to be programmed and trained with data and human instructions. In this case, speech pathologists had listened to voice recordings of people with HD counting from 1-20 forward and backwards, manually marking or labeling features of their speech and sounds. Bachoud-Levi’s team used this data to train xHD-Vox to identify and categorize features of speech, and xHD-Vox pulled out a set of important features. This included deficits in articulation and sound, changes in the rhythm and speed of counting, as well as pauses, the addition of extra sounds, and errors in the sequence of the numbers. So basically, they were teaching the computer device to rate and assess the patients’ speech recordings, without the need of a human evaluator.

The authors then showed that the computer-rated vocal features were strongly correlated with  performance on clinical tests, in particular, the gold standard cUHDRS, essentially making it a digital “biomarkerbiomarker a test of any kind – including blood tests, thinking tests and brain scans – that can measure or predict the progression of a disease like HD. Biomarkers may make clinical trials of new drugs quicker and more reliable.”.  This was a crucial step, as it gave the authors more confidence that this voice test was accurately reflecting underlying HD biology.

In the next part of the study, the researchers tested how well xHD-Vox could track HD progression over a 2-year time frame.  What they found was that xHD-Vox could predict the decline in HD symptoms in a manner similar to what the clinicians could predict using clinical scores.  One very exciting part of this study is that their model can be implemented into mobile apps that could be used on a person’s smart phone for remote monitoring of their own voice over time.

What are the next steps?

Using artificial intelligence in this manner is ground-breaking. However, xHD-Vox is not quite ready for prime time. Before it can be considered for wide-spread use, it will be important to test xHD-Vox in additional populations of people with HD and test real-world voice recordings using mobile phones. With this in mind, the researchers have already developed an Android prototype of xHD-Vox to explore its potential for mobile deployment and remote monitoring. This is not something you’ll find in your app store, but it’s currently being tested experimentally.

Additionally, this device needs to be tested in other languages, besides English. A truly effective biomarkerbiomarker a test of any kind – including blood tests, thinking tests and brain scans – that can measure or predict the progression of a disease like HD. Biomarkers may make clinical trials of new drugs quicker and more reliable. needs to be globally accessible. There has only been one multilingual voice analysis study in HD (Fahed et al) which focused on reading tasks, and this study showed that reading features were language-dependent. With Whisper supporting 96 languages, the researchers believe that adapting xHD-Vox to a new language would be easy and efficient because it involves counting numbers versus reading words.

Nonetheless, this work represents an important step forward in telemedicine for HD monitoring, with the potential to detect disease progression. This could be especially important in regions with limited medical access or rural communities.

Summary:

• Human speech depends on precise coordination between the brain, vocal cords, tongue, and lips.
• Scientists study these connections to understand how the brain produces speech—and how this process changes in diseases like HD.
• People with HD may speak more slowly, pause more often, or have difficulty coordinating the sounds needed for clear speech.
• Researchers developed xHD-Vox, an AI-powered tool that analyzes speech during a simple counting task.
• Speech features measured by xHD-Vox reflected clinical assessments and tracked changes in HD over two years.
• With further testing, voice analysis could offer a simple, noninvasive way to monitor HD remotely.