Restoration of noisy audio recordings seeks minimum degradation of sound and maximum suppression of noise. Spectral suppression methods perform best with high frequency resolution but the latter results in poor performance with transients. While Wavelet based algorithms attempt to optimize the time-frequency tradeoff, they suffer from frequency aliasing. The suggested pyramid algorithm is a good candidate to optimize the time-frequency resolution trade-off while avoiding aliasing. In this study an algorithm for removal of wide-band noise from old audio recordings is evaluated. The algorithm is based on the pyramid algorithm and on a spectral method for noise suppression. Results show enhanced conservation of onsets with efficient reduction of noise. The algorithm is implemented in real-time.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.