Psychoacoustic audio analysis in frequency domain
Résumé
Considering the linear frequency scale obtained when applying the Fourier Transform to a digital audio signal x[n], it is possible to perform an energy summation of the amplitude values related to x[n], based on the concept of Riemann sum, which is usually related with the mathematical definition of the definite integrals. It is presented the use of this technique for obtaining the root-mean-squared values (RMS values) as a result of a critical-band filtering of x[n]. Furthermore, it is implemented a dynamic amplitude weighting based on an interpolation of the equal-loudness-contours (Standard ISO 226:2003) using Artificial Neural Networks in an analogue procedure to the method defined in the Standard ISO 532:1975, part B. Examples of the application of these loudness models are shown and they are compared with the results obtained when using the A and C frequency weighting curves.