Spektrogram

A spektrogram, commonly called a spectrogram in English, is a visual depiction of how the spectral content of a signal changes over time. It represents the signal’s frequency content as a two-dimensional image where the horizontal axis denotes time, the vertical axis denotes frequency, and the intensity or color of each point encodes the magnitude of the spectrum at that time and frequency. Spectrograms are widely used to analyze audio signals, speech, music, and other time-varying signals.

Construction typically involves dividing the signal into short overlapping frames, applying a window function to each frame, transforming the windowed frames with the Fourier transform to obtain their magnitude spectra, and assembling these spectra as columns of a time-frequency matrix. The window length controls time and frequency resolution; common windows include Hann or Hamming. The magnitude is often displayed on a logarithmic scale to enhance visibility of weaker components, and the frequency axis can be linear or logarithmic depending on the application.

Interpretation and uses: In speech analysis, spectrograms reveal formants and phonetic transitions; in music, they show harmonic content, timbre, and rhythm; in bioacoustics they reveal animal vocalizations; in engineering they monitor machinery and other signals. Limitations include the inherent trade-off between time and frequency resolution due to the short-time Fourier transform, sensitivity to windowing, and the typical loss of phase information unless additional processing is performed.

Variants and related concepts: Mel spectrograms adjust the frequency axis to a perceptual scale; scalograms use wavelet transforms; other representations include power spectrograms and phase-aware forms. Modern workflows often use log-amplitude spectrograms with normalization and may derive features such as MFCCs from them.