The noise suppression algorithm is a speech enhancement algorithm based on spectrum estimation, primarily used to suppress environmental background noise and improve speech intelligibility. Its core idea is to convert the speech signal to the frequency domain using a short-time Fourier transform, estimate the noise power spectrum during frames without speech activity, and track changes in environmental noise through smooth updates.

During the suppression stage, the algorithm typically employs an improved Wiener Filter or spectral subtraction logic. By calculating the prior Signal-to-Noise Ratio (SNR) and the posterior SNR, the algorithm dynamically assigns gain factors to each frequency bin: deep attenuation is applied to frequency bands dominated by noise energy, while speech formant regions with high SNR are preserved. Additionally, a probability-based Speech Presence Probability (SPP) detection mechanism is introduced in the algorithm to perform nonlinear smoothing on the gain curve. Finally, the enhanced frequency-domain signal is reconstructed back to the time domain through an Inverse Fast Fourier Transform (IFFT).