To address the requirement of piecewise stationarity within the analyzed signal segments, today’s state of the art audio codecs make use of two filter bank resolutions. Short temporal resolution sequences are used to adapt to transient like jump signals, long temporal resolutions are used to effectively code the more steady or slowly drifting waveforms. With increasing computational capacity a better adaptation of the filter bank to the signal becomes feasible. The paper presents an experimental MDCT based transform coder which is capable of switching between four filter bank resolutions. A distortion measure is deployed which is driven by a simple psychoacoustic model that incorporates masking effects both for stationary and transient signals. A rate-distortion control is proposed to partition the signal to optimally match the signal contour with the temporal resolutions of the filter bank. Performance results are presented and compared to the conventional two resolution approach. Proposals for further developments, like pre-segmentation are evaluated.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.