Study Of Spectral Cues For Sound Source Localization Based On Multiple Datasets

The amplitude-frequency response characteristics generated by the interaction of sound waves and anatomical structures play an important role in auditory localization,especially in the median plane,so this amplitude-frequency characteristic is also called Spectral Cues in Sound Source Localization.The spectral cues can be decomposed into multiple peak-notch components of the amplitude spectrum,and the spectral notch is an important cue that affects the accuracy of localization in the median plane.In this study,the statistical analysis of the spectral peaks and spectral notches is carried out,and the threshold of the spectral notches position is measured by a psychoacoustic experiment.Finally,an objective model is used to verify and analyze the experiment.First,five head-related transfer function databases are taken as the original data.The attributes for component of the spectral cues from data within the median plane is extracted by using signal processing methods,and a statistical significance of the difference in elevation factors for the position attributes of spectral notches is performed.The results show that the positions of notches have a relatively consistent trend along with elevation angle,and the positions of notches in different elevation areas have different degrees of significant differences.It is preliminary judged that this kind of difference is related to the localization ability;the same operation is performed on the data in the horizontal plane,however,the results are lack of a similar trend.Then,sound signals is synthesized based on the above-mentioned statistical information of spectral cues,and a psychoacoustic experiment is performed for measuring the threshold for position of the spectral notch.The experiment uses the adaptive approximation three-downone-up method,and the response mode for the subjects is three intervals,three alternative forced-choice(3I3AFC),experimental variables are seven target elevation angles(-30 °,-15 °,0 °,15 °,30 °,45 °,60 °),each elevation angle contains two types of spectral notches,and there are left and right two approach directions for each type of spectral notches.The results show that the target frequency of the spectral notches has a significant effect on the threshold,and the approach direction factor has a slight effect on the threshold.The average threshold of the first spectral notch is about 0.5 k Hz,and the average threshold of the second notch is about 1 k Hz,both of which have a tendency to rise as the center frequency rises.At the same time,it finds that this threshold result can correspond to the average discrimination of the localization for elevation in this area,indicating that the localization for elevation is carried out to some extent by relying on the change of the location of the notch structure.Finally,an auditory difference evaluation model is used to objectively analyze the above threshold experiment for spectral notches shifting.The results show that the difference evaluations for shifting from most of the target frequency group have good left-right symmetry;If taking a specific ODG value as the observation position,the intervals between left and right shifting of both the first and second spectral notches increase as the target frequency position increases,the extent of shifting for the second spectral notch part is overall larger than the extent for the first spectral notch part.The results of objective analysis are basically consistent with the experimental results,which proves the validity of the experimental data.This study helps to further understand the principle and mechanism of median plane localization,and also provides a reference and basis for the customization and simplification of the auditory transfer function.