Persistent Time-Frequency Shrinkage for Audio Denoising
In many audio processing applications, signals are represented by linear combinations of basis functions (such as with windowed Fourier transforms) that are collected in so-called dictionaries. These are considered well adapted to a particular class of signals if they lead to sparse representations, meaning only a small number of basis functions are required for good approximation of signals. Most natural signals have strong inherent structures, such as harmonics and transients, a fact that can be used for adapting audio processing algorithms. This paper considers the audio-denoising problem from the perspective of structured sparse representation. A generalized thresholding scheme is presented from which simple audio-denoising operators are derived. They perform equally well compared to state-of-the-art methods while featuring significantly less computational costs.
Click to purchase paper or login as an AES member. If your company or school subscribes to the AES Journal then you can look for this paper in the institutional version of the Online Journal. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $20 for non-members, $5 for AES members and is free for E-Library subscribers.