The quality of audio recordings is often degraded by various types of disturbances, such as broadband noise, hum, clicks, and crackles. Of these, broadband noise is one of the most frequently occurring types of disturbance, especially in old recordings. Disturbances can be classified as having either a technical or acoustic origin. This research presents a novel algorithm to estimate the power spectral density (PSD) of stationary broadband noise disturbances in audio recordings. The proposed algorithm estimates the noise PSD as the mean value of an exponential distribution that corresponds to the truncated periodogram coefficients of the disturbed audio signal. A confidence value is computed to reflect the reliability of the noise PSD estimate. Noise PSD estimates with a low confidence are rejected in order to avoid degrading the desired signal when the obtained noise PSD estimate is used in a noise-reduction algorithm. Based on experiments with a large database of clean speech and music signals and different artificial and real-world broadband noise disturbances, the results show that the proposed algorithm yields reduced PSD estimation errors compared to the state-of-the-art minimum statistics algorithm for a large range of SNRs. The algorithm allows for unsupervised operation and thus constitutes an important part of a fully automatic broadband noise restoration system for audio archives.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.