Perceptual evaluation of interior aircraft sound models

We report a listening test conducted to investigate the validity of sinusoids+noise synthesis models for interior aircraft sounds. Two models were evaluated, one for monaural signals and the other for binaural signals. A parameter common to both models is the size of the analysis/synthesis window. This size determines the computation cost and the time/frequency resolution of the synthesis. To evaluate the perceptual impact of reducing the window size, we varied systematically the size N_s of the analysis/synthesis window. We used three reference sounds corresponding to three different rows. Twenty-two participants completed an ABX discrimination task comparing original recorded sounds to various resynthesized versions. The results highlight a better discrimination between resynthesized sounds and original recorded sounds for the monaural model than for the binaural model and for a window size of 128 samples than for larger window sizes. We also observed a significant effect of row on discrimination. An analysis/synthesis window size N_s of 1024 samples seems to be sufficient to synthesize binaural sounds which are indistinguishable from original sounds; but for monaural sounds, a window size of 2048 samples is needed to resynthesize original sounds with no perceptible difference.