Human hearing is wired so that the brain prioritizes familiar patterns — a friend's voice, the patter of rain, a car horn. But a new system works on a different principle: it registers any sound burst, even one never encountered before. This approach is called audio diarization: it's not just about finding specific noises but segmenting an entire audio recording into fragments with clear start and end markers for each sound event. Unlike classical recognizers, this method doesn't require a pre-defined list of known sounds. You could compare it to a vigilant sentinel who reacts to the slightest movement without asking whether it's a friend or a random passerby.

Technically, the method relies on the same idea as spectroscopy: a complex sound is decomposed into simple harmonic waves, much like white light splits into a rainbow of colors. This mathematical possibility was discovered two centuries ago by Joseph Fourier, who showed that any oscillation can be represented as a sum of sinusoids. The system applies the Fourier transform to the incoming audio stream and obtains a spectrogram — the distribution of energy across frequencies at each moment in time. Then entropy comes into play — a measure of surprise, or informativeness, of the signal. Segments of the recording with a sharp rise in entropy are interpreted as boundaries of sound events, because that's where the statistical structure of the signal changes: background noise gives way to something new. In this way, the algorithm separates important signals from the monotonous background without needing libraries of samples.

Such an "all-hearing" algorithm could be useful in smart homes to notice equipment malfunctions (for example, an unusual knock from a refrigerator compressor) before the failure becomes obvious. In the wild, it would enable animal monitoring by registering any vocalizations without prior training on specific species. And it's a step toward creating assistants for people with hearing impairments: what truly matters is not just recognizing the familiar but also catching an unfamiliar sound in time — a danger signal or simply an unexpected event that a hearing person would notice effortlessly.