Article
Hearing Equivalent Signal Analysis by Auditory Images in Industrial Applications (de)
* Presenting author
Abstract:
The discrete short-time Fourier transformation is usually used for the analysis of time-discrete acoustic signals. It describes the physical signal properties in the frequency domain but does not consider the physiology of the human ear, although frequency weighting of the amplitude spectrum is commonly performed, and the frequency axis is represented logarithmically.The cochlea nucleus is considered as an early processing stage in human hearing where e. g., onset and pitch detection, periodicity perception or binaural hearing take place. Based on a cochlear model (Lyon, 2017) the theory of auditory images (Licklider, 1951; Patterson et al., 1992) provides a representation of sound according to the first stage of the human hearing sense.We have successfully used auditory images in several use cases in which conventional analysis methods have not delivered clear results. Auditory images have the advantage that they contain all the information about a sound; phenomena such as roughness, fluctuation, periodicity, pitch (= tonality), just to name a few, can be found directly. This universal, aurally accurate representation is supposed to deliver better results in machine learning applications.