Psychoacoustic Overview

Due to their physiology, the outer, middle and inner ear don't respond linearly to pressure waves depending on their incidence, frequency or amplitude. This behaviour can be designed to ease sound sources detection, localisation, or protect the auditory system. More generally, this affects pitch, intensity and timbre perception, which is not linear.

When a cell is excited by the pressure wave, it releases neurotransmitters at the same frequency as that of the wave. Consequently, we can also code the pitch temporally, and not only from the location of the excited cells. This is the case for instance when we perceive the pitch of a fundamental from the frequency of the upper harmonics of a sound, even if no energy is found in the actual fundamental of a sound. This is a subjective, or virtual perception of pitch.

This means that the pitch is detected with two different mechanisms : tonotopic and temporal coding. The coexistence of these mechanisms is especially noticeable when listening to sounds with very low and/or weak fundamental frequencies, with resonances within the dominant audible area. The  ear is not very with sensitive below 200 Hz, the ear is less sensitive, and we have trouble perceiving pitches below 30 Hz. Our ear is very sensitive to frequencies between 300 and 8000 Hz. We can detect this fundamental frequency, but we can also focus on the pitch of these higher resonances. This is the case with diphonic chant for instance, or with the extreme register of low instruments, for which the ear is not very sensitive, or cannot even detect a clear tone – below 30 Hz.

The tuning of the basilar membraneis also called auditory filtering, or critical bandwidth. Each critical bandwidth has a frequency resolution. This resolution determines the ability to choose a filter which is centered over the expected frequency of the signal, but no other frequency.

If a signal is masked by a different frequency, this means the auditory system can't distinguish between the two frequencies. This is the case when two frequencies are similar, or quite close, when we listen to a voice withing the background noise of the street... Experimentations about masking conditions allow to determine the frequency selectivity of the auditory system.

The greatest masking is when the masker and the signal are in the same auditory filter, that is, have the same frequency.

When two frequencies are quite close, a frequency can mask neighbour frequencies, if its amplitude is higher. The masking is especially significant when a low frequency is used to mask a higher frequency. The effect grows with the amplitude of the signal.

The Terhardt's virtual pitch algorithm is designed to recalculate the amplitude spectrum integrating the processing of spectral pitch by the brain, and includes a number of psychoacoustic effects, among which sensibility threshold variations, multiple complex tones that which evoke several pitches, and masking effects.

The Fourier analysis of a signal yields approximation patterns of frequencies and amplitudes. The algorithm translates these patterns into spectral-virtual-pitch patterns according to

• hearing threshold

• masking effects

• components interaction in subjective pitch

• components relative weight and frequency

• formation of virtual pitches by subharmonic coincidence.

Source : http://www.uni-graz.at/richard.parncutt/ptp2svpDoc.html

1. You can possibly select the region you want to analyse, with a click and drag.

2. Choose Analysis/Masking Effect.

3. In the dialogue window, specify the analysis parameters.

4. Press OK

In the dialogue window, specifiy a name and location for the file.

• Maximum Number of Peaks : maximum number of peaks in the output. The peaks selected are the peaks with the largest amplitude.

• Threshold : attenuation below the peak amplitude. A peak is taken if its maximum is at least equal to this threshold above the neighboring minima.

  For instance, if the number of peaks is 20, with a 30 dB threshold, the 20 strongest peaks with an amplitude larger than 30dB above the two neighboring minima will be detected.