spektrinsubtraktion

Spektrinsubtraktion, also known as spectral subtraction, is a signal processing technique used to reduce noise in audio signals. The core idea is to estimate the noise spectrum during a period when the desired signal is absent and then subtract this estimated noise spectrum from the entire signal. This process aims to preserve the desired signal components while attenuating the noise.

The algorithm typically involves identifying a segment of the audio that is believed to contain only noise. The Fast Fourier Transform (FFT) is applied to this noise segment to obtain its spectral characteristics. This noise spectrum is then averaged to create a noise profile. Subsequently, the FFT is applied to the entire signal, and at each frequency bin, the magnitude of the estimated noise spectrum is subtracted from the magnitude of the signal spectrum. The phase information of the original signal is usually preserved.

Spektrinsubtraktion is effective for stationary or slowly varying noise, such as background hum or hiss. However, it can introduce artifacts, like "musical noise," which manifests as tonal artifacts or ringing in the processed signal, especially when the subtraction is too aggressive or the noise is not stationary. Various modifications and refinements to the basic spektrinsubtraktion algorithm exist to mitigate these artifacts and improve performance in more complex noise scenarios. It finds applications in areas like speech enhancement, audio restoration, and noise reduction in various audio recording and playback systems.