Real-time performance measures of perceptual audio coding

The selection of the wavelet packet best basis of a perceptual real-time audio encoder has been a challenging task for researchers. Due to the subjective nature of the auditory masking phenomenon, fully objective performance measures (entropy versus SNR, delay, etc.) are not sufficient for the selection of optimal representations. In our work we present a DSP application by means of which we can measure, for each hybrid wavelet packet basis resulting from a predetermined library of wavelet filters, the bitrate of a perceptually transparent coding of the audio input and the overall filter bank delay. The novelty is that the transparency of the compression is established with a subjective evaluation. The tool user can customize both the filter bank parameters and the psycho-acoustic model parameters by means of a MATLAB/spl reg/ graphical interface and listen in real-time the reconstructed signal. The audio codec runs in real-time on a TI TMS320C6711/spl trade/ DSK.