Research & AlgorithmsDANIEL SCHMITTER · RESEARCH & ENGINEERING

Practical sensing · Research & Algorithms

Can a laptop hear you breathe?

A speaker, a microphone, and an operator that recognizes a tone: exploring how tiny changes in reflected sound might reveal slow motion.

Daniel Schmitter · Research story · Revised draft

Read the technical report ↗
EXPLORE THE IDEA

The room becomes part of the sensor

A reflected acoustic path carries tiny changes in motion.

THE MEASUREMENT PATHSound probes a moving reflectorspeakermicrophonereflectorSchematic path; no human or measured motionOPERATOR-INFORMED FRONT ENDCarrier → local dynamics → slow cue20 kHz carrier · 48 kHz samplingSynthetic modulation; no physiological label
50%
Original acoustic-path diagram and computed synthetic modulation. This is an audio motion prototype, not Wi-Fi or validated respiratory measurement. The archive used a 20 kHz carrier and 48 kHz audio sampling. No sound plays.

Follow the information

From input to outcome

The physical path comes before the estimator. Motion alters a mixture of acoustic paths; the recurrence-based calculation extracts a motion-sensitive feature, not calibrated chest displacement or a clinically validated breathing rate.

Scroll the diagram horizontally to follow the route. Keyboard: focus the diagram, then use the arrow keys.

Speaker carrier → Room + moving reflector → Microphone samples → Narrowband estimation → Slow motion proxy. The physical path comes before the estimator. Motion alters a mixture of acoustic paths; the recurrence-based calculation extracts a motion-sensitive feature, not calibrated chest displacement or a clinically validated breathing rate.
Information-flow map. Audio, not Wi-Fi. The recording lacks a synchronized respiratory reference. Original vector schematic based on the method and evidence discussed in this article; signal shapes and icons are illustrative, not additional measurements. Open full-size diagram ↗

Read the main route from left to right; labelled side branches show additional inputs, checks or feedback. The sections below explain the operations and their experimental limits.

The signal is already in the room

A laptop is usually a device for listening to people. Could its audio hardware also detect how a nearby surface moves? A transmitted tone returns to the microphone through several acoustic paths. Motion changes those paths, leaving small variations in a signal dominated by the tone and the room. Breathing is a compelling possible application because chest motion is slow and repetitive. The immediate research problem, however, is separating a useful motion signal from everything else the microphone hears.

Inside the acoustic measurement chain

The experiment is audio-based: a speaker emits a high-frequency carrier and a microphone records a mixture of direct sound and reflections. Slow motion changes those paths. Narrowband filtering isolates the carrier region; a short-window dynamical estimator extracts a motion-sensitive quantity; subsequent filtering exposes slower variation. It is not a Wi-Fi channel measurement.

A local single-tone recurrence supplies the mathematical model, but the room supplies multipath, device filtering, amplitude changes, and unrelated movement. The estimator therefore does not directly output chest displacement. The approximately 8.1-cycle-per-minute peak in one recording is a candidate slow modulation, not a validated respiratory-rate label.

The information and computation flow of this example
Explicit component and information boundaries. Original scientific diagram; the stated component and information flow, not an additional experiment. Open full-size figure ↗

Recognize the carrier through its dynamics

A pure sampled sinusoid obeys a short recurrence. Three consecutive samples are related by a coefficient determined by frequency. That is the discrete counterpart of the oscillator differential operator, and it gives an inexpensive way to test whether a short segment behaves like a tone. The operator viewpoint replaces an arbitrary collection of signal features with a model of how the carrier evolves.

xn+1−2cos⁡(ωΔt)xn+xn−1=0\begin{gathered}x_{n+1}-2\cos(\omega\Delta t)x_n+x_{n-1}=0\end{gathered}
A single ideal sinusoid satisfies this relation exactly. Reflections, frequency changes, and noise make the residual informative—but also harder to interpret.

From a recurrence to an estimator

For a narrowband, approximately stationary segment, a least-squares estimate of the recurrence coefficient can be formed from dot products. Mapping that coefficient through an arccosine gives a frequency estimate under the single-tone model. The real acoustic scene is less cooperative: direct sound, multiple reflections, amplitude changes, microphone processing, and noise all affect the estimate. A changing frequency estimate is therefore a motion-sensitive feature—not automatically a calibrated chest-displacement or respiratory measurement.

What the prototype actually did

The recorded prototype used a 20 kHz carrier with 48 kHz sampling and speaker–microphone hardware. It isolated a narrow carrier band, estimated short-window dynamics, and used filtering and accumulation to expose slower variation. The implementation used 50 ms context; a smaller nominal step parameter does not establish an independent new measurement every 10 ms. The distinction matters when translating a signal-processing script into a latency claim.

A promising trace is not yet a breathing-rate measurement

One roughly 44-second recording contains a low-frequency peak near 0.134 Hz, or about 8.1 cycles per minute. That is consistent with a possible respiratory timescale, but the recording does not include an independently synchronized respiratory reference. We therefore cannot identify the peak as validated breathing or report clinical accuracy. The animation shows the mechanism we want to test; it does not fill in the missing ground truth.

Why this belongs in a machine-learning research series

The interesting bridge is an operator-informed front end: a small dynamical calculation could supply meaningful features to a learned estimator, or reduce the learning burden by removing a known carrier. The prototype does not establish a trained respiratory model or a spline-specific performance advantage. Its value is a concrete, inexpensive experimental setting where representation, signal extraction, noise, and eventual learning have to work together.

Turn a motion cue into a validated sensing experiment

The compelling next interface would connect this operator-informed front end to synchronized physiological ground truth and ordinary demodulation controls. No person, chest-motion video, or artificial waveform should stand in for that missing validation. The revised visual concentrates on the actual speaker–reflection–microphone path and explicitly synthetic modulation, without suggesting a medical measurement was established.

The experiment that would make the story stronger

A synchronized reference belt or another validated respiratory reference would let us compare estimated motion with actual breathing. A motionless reflector, an empty room, ordinary demodulation, and deliberate non-respiratory movement would test alternative explanations. Those controls could turn a compelling physical mechanism into evidence for a useful sensor. Until then, this is an acoustic motion-sensing prototype—not a medical device.

Evidence & further reading

The links below distinguish the project record from foundational literature. This revised story does not add a new application-validation experiment.

  1. Contactless acoustic motion-sensing report. Local signal-sensing project (2026). Local archive snapshot.
  2. Contactless acoustic motion prototype: ghost_v2. Local signal-sensing project (2026). Local archive snapshot.
  3. Flagship stories, source corrections and visual direction. Publication audit (2026). Local archive snapshot.

Other posts of interest