The Sun is not just a blazing ball, but a colossal orchestra, where the strings are solar filaments — giant arches of dense plasma floating in the scorching corona. They tremble, and each oscillation is an encrypted melody of magnetic fields. Previously, astronomers manually ‘listened’ to the images: placing virtual slits, building time-distance diagrams, searching for repeating signals. Titanic labor, meager catch: in two weeks — a couple of dozen events. Now, a listening algorithm steps onto the stage, picking up four times as many.

The pipeline’s heart is the data from GONG, a network that daily images the Sun in the Hα line of atomic hydrogen (the famous Balmer series). First, photometric correction removes limb darkening and evens out brightness. Then machine learning steps in: convolutional networks automatically segment the filaments, extracting regions of interest. For each, a multiscale representation is built: images are averaged with kernels of different sizes, and for each cell a Lomb–Scargle periodogram is computed — a gap-tolerant method based on Fourier transform. Thus a spectral map of vibrations is born.

But how to distinguish true plasma tremors from random fluctuations? The key is adaptive threshold calibration. Background noise is modeled by a convolutional network trained on numerically simulated spectra. Then for each day, a method inspired by conformal prediction computes a correction curve that adjusts the threshold to real data. Anything exceeding δ=6×10⁻⁷ (roughly five standard deviations) and appearing on at least four scales is considered a reliable event.

The result is staggering: in the first two weeks of January 2014, the pipeline identified 91 oscillatory events with periods from 20 to 126 minutes, while the manual catalog contained only 22. Individual matches are confirmed, but the main thing is a whole population of previously unnoticed vibrations has been uncovered. The solar corona is literally shot through with tremors, and we are only beginning to hear it. And who knows — perhaps in these vibrations lies the answer to why the corona is hundreds of times hotter than the surface of our star: after all, oscillations are directly linked to energy transport along magnetic loops.

Now, armed with automatic hearing, physicists will move to large-scale prominence seismology: from the period and damping of oscillations they reconstruct the geometry and strength of magnetic fields, and in the future — track how filament dynamics change over the 11-year cycle. Each oscillation is not just a note in a symphony, but a possible precursor to eruptive events: coronal mass ejections and flares capable of striking Earth. To hear the tremors in time is to take a step toward reliable space weather forecasting.