Spectral Windowing for Enhanced Temporal Noise Shaping Analysis in Transform Audio Codecs

Temporal Noise Shaping (TNS) is a method employed in many transform audio codecs. It works by performing open-loop linear predictive coding (LPC) in the frequency domain. In state-of-the-art codecs the Autocorrelation Function (ACF) required for the determination of the LPC coefficients during TNS anlysis is calculated on a spectral region cut out with a de-facto rectangular window. In this work a detailed mathematical description of the TNS analysis is done to show that rectangular spectral windowing leads to a suboptimal temporal resolution of TNS. Furthermore a new windowing scheme is proposed that allows for a good tradeoff between temporal and spectral resolution for TNS. The effect on the perceptual quality for different variants of this scheme is evaluated with a listening test in an experimental coding environment.