Your Heartbeat Could Be Your Password—And a Laser Can Read It Through Your Skin

You set down your coffee cup at a cafe. A smudge of your fingerprint remains on the glass. Someone photographs it, prints a replica, and now has access to your phone. Or consider this scenario. You post a selfie online, and facial recognition software maps every contour of your face—data that could be used without your knowledge.

What if there were a form of identification hidden inside your body that others couldn’t take or copy? What if your own heartbeat became the key to your digital life?

A group of researchers at the University of Glasgow has built a system that does exactly this very thing. Using laser light and camera technology, they capture heart sounds through the skin and process these acoustic signatures using brain-inspired computing. The approach achieved 89.03% accuracy in identifying twelve test subjects (meaning it correctly recognized nearly nine out of ten people). This offers a glimpse at what biometric security might look like in the future.

What Makes This Different

This is the first display of truly contactless, laser-based biometric identification using heart sounds processed by hardware neuromorphic computing. The system combines privacy, convenience, and power savings without relying on cloud servers or direct physical contact.

Safety note: The laser used is eye-safe, operating at power levels below 0.5 mW (about as strong as a typical laser pointer)—gentle enough for long-term skin exposure.

Privacy advantage: All processing happens on your device. Your cardiac signature never travels to a server or cloud, keeping your biometric data entirely under your control.

How Heart Sounds Stack Up Against Other Biometrics

Before diving into the technology, here’s how this approach compares to existing methods:

MethodForgery RiskHygieneNo-ContactCloud-FreeAccuracy (Study)
PasswordHighYesYesVariesUser-dependent
FingerprintMed-HighNoNoSometimesHigh
Face IDMediumYesYesSometimesHigh
Heart-LaserVery LowYesYesYes~89%

Key takeaway: Heart-based identification is internal, making it very difficult for anyone else to copy or steal. The combination of contactless operation, difficult to forge signatures, and local processing makes cardiac biometrics particularly appealing for high-security applications.

Why Hearts Tell Better Stories Than Faces

The human heart produces sounds with each contraction. This is the familiar “lub-dub” that doctors listen for with stethoscopes. These sounds come from valve closures and blood flow patterns. Every person’s heart creates a slightly different acoustic signature because of variations in heart structure, valve function, and even skin thickness.

Think about your daily routine. You unlock your phone dozens of times. Each instance potentially exposes your face to cameras or your fingerprint to surfaces. Fingerprints and facial images can be acquired and duplicated, leading to privacy breaches and fraud. You can’t leave your heartbeat on a drinking glass. No one can photograph it from across a room. The internal nature of cardiac sounds provides an inherent layer of protection that external biometrics lack.

The research team’s optical stethoscope works by directing an eye-safe green laser at the neck while a high-speed camera captures the reflected light patterns. Here’s what happens:

When laser light hits skin that’s vibrating from your heartbeat, it creates interference patterns called speckle. Think of the tiny shimmering dots you see when sunlight reflects off a CD or the way light dances on rippling water. As your heart beats, these dots shift position. By tracking these shifts using optical flow algorithms (software that detects motion between video frames, like how your phone’s camera tracks your face), the system reconstructs your heart sound without any physical contact.

During testing, participants sat naturally in an office chair while the laser-camera system recorded their neck vibrations for 10 second intervals. The camera captured frames at 1.47 kHz, about 1,470 pictures per second. Each frame showed subtle pixel displacements caused by blood flow driven by cardiac contractions.

What This Feels Like in Real Life

At the airport: Instead of fumbling with boarding passes or staring into a camera while trying not to blink, you sit in a booth for ten seconds. A small green light, barely noticeable, points at your neck. No scanner touches your skin. No one asks you to remove your glasses or adjust your posture. Your unique cardiac signature confirms your identity, and you’re cleared to proceed.

Unlocking your car: After a workout, hands full of gym gear and face flushed, your phone reads your elevated but still uniquely yours heartbeat through your shirt. The door opens. No fumbling for keys, no smudged fingerprint readers failing to recognize your sweaty finger.

Smart home entry: Arriving home with grocery bags in both arms, you pause at the door. A small sensor mounted in the doorframe reads your heartbeat through your clothing. The lock disengages. You walk in without setting anything down.

Morning phone check: You pick up your phone fresh from the shower, face still wet, hair uncombed. The embedded laser module captures your cardiac rhythm through your palm as you hold the device. It unlocks instantly—no need to dry your face or position yourself perfectly in the camera frame.

This isn’t a fictional movie scene, it’s the practical application the Glasgow team envisions for their technology.

Tech Details: Computing Like a Brain

Casual readers can skip this section and jump to “Testing Under Real Conditions”

Processing these heart signals required a departure from conventional machine learning. Standard neural networks run on computers where memory and processing units are separate—a setup that consumes considerable power shuttling data back and forth. The Glasgow team instead turned to physical reservoir computing, specifically a model called rotating neuron reservoir (RNR).

Think of it this way: A typical computer is like a library where you have to walk between the shelves (memory) and a reading desk (processor) every time you need information. Physical reservoir computing is more like having everything you need spread out on one big table where you can see and use it all at once.

This hardware-based neuromorphic architecture keeps signals in analog form (continuous waves, like sound waves on an oscilloscope) throughout processing. It combines memory with computation in the same physical components. Software-based approaches convert everything to digital bits (ones and zeros) and process them separately. The RNR system processes information directly as it arrives, similar to how ripples in a pond store information about multiple stones thrown into the water. Each ripple pattern overlaps and creates a unique signature that the system can “read.”

The reservoir consists of interconnected neurons arranged in a loop. As signals flow through this network, neurons (built from resistor-capacitor circuits and diodes—types of basic electronic parts found in radios and computers) pass information forward while also remembering what just happened. The connections rotate at each time step using switches called multiplexers (electronic switches that route signals to different paths), creating a dynamic system well-suited for recognizing patterns that change over time.

What makes this approach particularly practical is its training simplicity. While the input and reservoir layers stay fixed, only the output layer requires training through linear regression (a straightforward mathematical technique for finding patterns in data). In other words this is not the computationally intensive methods used in deep learning networks.

Signal preprocessing: The system cleans raw video data by applying a bandpass filter (only lets through sounds within a certain range, like tuning a radio to one station) with cutoff frequencies of 20 and 700 Hz to remove unwanted wobbles from skin or neck movement. The filtered signals are then resampled at 147 Hz to match what the network can effectively remember. The dataset labeling takes a continuous, point-by-point approach, with markers expanded to neighboring positions to reduce sensitivity to manual labeling errors.

Testing Under Real Conditions

Laboratory results often paint an optimistic picture that crumbles in real-world use. The research team put their system through several practical scenarios to gauge its strength.

After exercise: Heart sound data was collected from eight individuals following two-minute jumping jack sessions. Your heart beats faster and sounds different after exertion. The system was trained to recognize both calm and elevated heart patterns. It first checks whether your heart rate is above 100 beats per minute (elevated) or below (normal), then routes the signal to the appropriate trained reservoir for identification. This mixed dataset yielded 83.01% accuracy (correctly identifying about eight out of ten people even when their hearts were racing).

With heart conditions: The researchers evaluated performance with individuals who have heart murmurs or other cardiac irregularities. Using publicly available datasets, they simulated heart murmurs by playing audio through a speaker covered with foil while capturing the reflected laser speckles. When four types of heart murmurs were added to the original dataset, the system maintained approximately 90% accuracy in distinguishing sixteen people (correctly identifying nine out of ten individuals). A real abnormal heart sound from a colleague with cardiac problems was also successfully incorporated, achieving 89.81% accuracy (nearly nine out of ten correct identifications).

These results suggest that cardiac conditions don’t prevent identification but rather simply add to the unique signature of each individual’s heart.

The Hardware Advantage

Comparing the RNR system to established machine learning revealed interesting insights. Against a support vector machine (SVM) classifier using 16 time-domain features and a bidirectional LSTM network (a type of artificial intelligence commonly used for pattern recognition), the RNR achieved 89.03% accuracy—slightly below the LSTM’s 90.56% but above the SVM’s 85.42%.

What these numbers mean: In tests, the system correctly identified nearly nine out of ten people, showing it can reliably tell one heartbeat from another.

The key difference lies not in raw accuracy but in how computation happens. The SVM and LSTM require complex training algorithms, signal segmentation, and feature extraction that consume substantial computational resources. The RNR processes raw signals directly with simple linear regression training—no need to pre-analyze or break apart the data first.

Breaking down the performance metrics more granularly: The RNR system achieved sensitivity of 89.3% (correctly identifying the right person 89 times out of 100), specificity of 99% (rarely confusing one person for another—only 1 mistake per 100 tests), precision of 89.31%, and an F1 score of 88.87% (a balanced measure that accounts for both accuracy and reliability). These numbers indicate balanced performance across true positive and true negative classifications, with particularly strong specificity suggesting the system rarely misidentifies someone as the wrong person.

Memory capacity analysis showed the network could remember information from 68.91 time steps in the past, sufficient for recognizing patterns across multiple heartbeats. The team found that increasing reservoir size improved performance when identifying larger numbers of individuals, ultimately settling on 6 parallel reservoirs of 500 neurons each (3,000 neurons total) for the 12-person classification task.

Noise analysis tested the system’s resilience to real-world imperfections. By artificially adding random static to collected heart sounds, researchers found the system maintained high classification accuracy with signal-to-noise ratios above 20 dB (about the difference between a quiet library and normal conversation)—achievable with commercial laser and data acquisition equipment. This means the technology works even when conditions aren’t perfect.

From Lab to Pocket

The practical vision for this technology extends beyond research laboratories. Smartphones and smartwatches could integrate small laser modules to collect heart sound data, with an RNR processor serving as the processing core. Such a setup could even be deployed in photo booths, allowing continuous identity monitoring or health checks without requiring contact with sensors.

Several factors make optical stethoscope measurements particularly suitable for widespread adoption:

Works in crowds: The focused nature of laser sensing makes it robust against interference from other moving humans nearby. Radar-based systems need you to be alone in a room. The laser-camera approach functions in busy settings as long as the subject remains visible.

Noise-proof: It works in noisy environments where voice recognition would fail—think construction sites, subway platforms, or busy restaurants.

Hygiene and convenience: The non-contact nature satisfies both hygiene concerns and user convenience. Participants in the study could move their limbs freely during recording, a stark contrast to systems requiring precise finger placement or stillness.

Data stays local: Processing occurs on the device rather than transmitting sensitive biometric data to cloud servers, reducing bandwidth needs while improving response times and protecting privacy.

Everyday Impact: What This Means for You

Hygiene benefits: No more touching public fingerprint scanners that thousands of people have pressed their fingers against. No shared devices that need sanitizing between users.

Convenience wins: Access your accounts while cooking, exercising, or when your hands aren’t free. No need to stop what you’re doing to authenticate yourself.

Security upgrade: Your cardiac signature is internal and nearly impossible to steal or replicate. Unlike a password you might reuse or a face that appears in photos, your heartbeat stays private.

What Could Go Wrong?

No technology arrives without questions that need answering. Several concerns warrant consideration before this system moves from laboratory to marketplace.

Cardiac variability: While the study successfully identified individuals with various heart murmurs and one person with cardiac problems, larger-scale validation is needed. How does the system perform with severe arrhythmias, valve replacements, or pacemakers? These devices and conditions might alter cardiac signatures unpredictably.

Public perception: The green laser diode operates at a wavelength of 532nm with power exposure below 0.5 mW on human skin, considered safe for long-term eye and skin exposure. However, public perception of lasers pointed at the body may create adoption barriers, regardless of safety data. Many people feel uneasy about lasers near their skin or eyes, even when told they’re harmless.

Spoofing attempts: While heart sounds are difficult to forge, determined adversaries might attempt to play recorded cardiac audio near the sensor or use sophisticated vibration devices. The system’s resilience to such attacks requires further testing.

Physiological diversity: The current study involved participants aged 22 to 32. Performance across broader age ranges, body types, skin tones, and neck anatomies needs validation. Skin thickness, melanin content, and subcutaneous tissue composition could all affect laser speckle patterns.

Movement and positioning: While participants could move their limbs freely, even small changes in head position or neck tension might alter readings. Real-world scenarios—checking your phone while walking, for instance—present more complex motion artifacts than sitting still in an office chair.

These concerns require resolution through expanded trials before this technology transitions from laboratory curiosity to practical security tool.

How Soon Could This Appear?

The Glasgow team acknowledges that full development remains ahead. Building larger, more functional public datasets across various conditions would help refine the system’s applicability. Chip-level integration of a 3,000-neuron network would require interdisciplinary collaboration to transition from laboratory prototype to consumer device.

Realistic timeline: Based on similar biometric technologies, consumer devices could see early versions within 3-5 years if development continues at current pace. Initial applications would likely appear in high-security environments (hospitals, government facilities) before expanding to consumer products.

What needs to happen first:

  • Larger studies with thousands of participants across diverse demographics
  • Miniaturization of laser and camera components for smartphone integration
  • Battery optimization to ensure all-day operation
  • Regulatory approval for medical-grade biometric devices
  • Public education campaigns about safety and privacy protections

Yet the foundation appears solid. Heart sounds provide a biometric that’s difficult to forge, capture, or abuse. Optical sensing enables non-contact measurement in diverse environments. And neuromorphic computing delivers processing that’s faster and more power-efficient than conventional approaches.

Join the Research

The next step in this technology’s journey is real-world trials and mass data collection. The University of Glasgow team seeks volunteers to help build more comprehensive datasets spanning diverse demographics, health conditions, and usage scenarios.

Who can participate: Adults of all ages, body types, and health conditions. Researchers particularly need data from:

  • People with various cardiac conditions
  • Different ethnic backgrounds and skin tones
  • Wide age ranges (from young adults to seniors)
  • Athletes and sedentary individuals

What participation involves: Sitting still for 10 second intervals while a laser-camera system captures your neck vibrations. Non-invasive, painless, and typically completed in under 30 minutes.

If you’re interested in contributing to the future of biometric security or simply curious about what your heartbeat sounds like through a laser, information about ongoing studies can be found through the University of Glasgow’s research portal.

As we keep searching for better ways to prove who we are, the answer may lie in the quiet rhythm that keeps us alive—a password encoded in every heartbeat, readable through skin, and recognized by devices that compute more like our own brains.

Citation: Ding, Y., Li, H., Liang, X., Vaskeviciute, M., Faccio, D., & Heidari, H. (2025). Physical reservoir computing for optical stethoscope-based heart sound biometric identification. IEEE Transactions on Artificial Intelligence. DOI: 10.1109/TAI.2025.3617382

Related Articles

PHP Code Snippets Powered By : XYZScripts.com
★

Track This Company

Create a free account to add companies to your watchlist and get notified when patents, funding, or key developments change.