A sound with changing time and frequency content.

Rippled sounds are dynamically changing sounds, and are therefore speech-like sounds. Actually, although ripples sound very artificial (similar tothe sounds used in old computer games), they are well-parametrized versions of speech. They are used in psychophysical testing to derive modulation transfer functions for normal-hearing and hearing-impaired subjects, and in auditory electrophysiology to derive spectral-temporal receptive fields for neurons.

The ampltiude of every frequency component changes as a function of time, which can be written in mathematical form:

\( S(t,x) = 1+\Delta M \cdot \cos{(2 \pi \omega t + 2 \pi \Omega x)} \)

with \( t \) time (s), \( x \) position of the spectral component in octaves above the lowest frequency, \( \omega \) ripple velocity (Hz), \( \Omega \) rippledensity (cycles\octave), and \( \Delta M \) the modulation depth on a linear scale between 0 and 1. In the next sections, the meaning of the parameters will be explained by generating a ripple in Matlab.

Matlab implementation

Let's take ripples consisting of a broadband complex of 128 components equally distributed (20/active) from 250 Hz to ~20 kHz.

F0      = 250; % base frequency (Hz)
nFreq   = 128; % (octaves)
FreqNr  = 0:1:nFreq-1; % every frequency step
Freq    = F0 * 2.^(FreqNr/20); % frequency vector (Hz)

All components had random phase, except for the first ( \( \phi_0 = 0 \) ).

Phi     = pi - 2*pi*rand(1,nFreq); % random phase
Phi(1)  = pi/2; % set first to 0.5*pi

Let's take a sound duration of 1000 ms, a modulation depth \( \Delta M \) of 1 (100%), a ripple velocity \( \omega \) of 4 Hz, and a ripple density \( \Omega \) of 0 octaves/cycle. We will change these parameters later, to see what the effects are.

vel   = 4; % omgea (Hz)
dens   = 0; % Omega (cyc/oct)
mod   = 100; % Percentage (0-100%)
durrip   = 1000; %msec
Fs    = 50000; % sample frequency (Hz)
nRip    = round( (durrip/1000)*Fs ); % # Samples for Rippled Noise
time  = ((1:nRip)-1)/Fs; % Time (sec)
Oct     = FreqNr/20;                   % octaves above the ground frequency
oct    = repmat(Oct',1,nTime); % Octave

The next step is to compute the dynamic amplitude modulations of the sound, according to the above formula.

%% Create amplitude modulations completely dynamic in a loop
A = NaN(nRip,nFreq); % always initialize a matrix
for ii = 1:nRip
  for jj = 1:nFreq
    A(ii,jj)      = 1 + mod*sin(2*pi*vel*time(ii) + 2*pi*dens*oct(jj));

Finally, we can add this envelope A to the carrier sound snd, and we will have our first ripple.

% Modulate carrier, in a for-loop
snd = 0;
for ii = 1:nFreq
  carr      = A(:,ii)'.*sin(2*pi* Freq(ii) .* time + Phi(ii));
  snd        = snd+carr;

When we look at the spectogram, we can see that the amplitude is changing only as a function of time. We have just created an amplitude-modulated (AM) sound.


This is exactly what happens if you change only the ripple velocity. Changing ripple density will lead to frequency-modulated (FM) noise. Changing both will lead to dynamic variations as shown in the top-figure. Try this out yourself. A negative density corresponds to an upward direction of the spectral envelopes, a positive density to a downward direction, and \( \Omega = 0\) indicates a pure amplitude modulated (AM) sound.

With the PANDA function pa_genripple it is easy to create these ripples and visualize them in a graph. The figure above can also be made with the following Matlab code.

t = (1:length(snd))/Fs;
ylabel('Amplitude (au)');
ylabel('Time (ms)');
xlim([min(t) max(t)]);
axis square;
box off;
nfft = 2^11;
window      = 2^7; % resolution
noverlap    = 2^5; % smoothing
cax = caxis;
caxis([0.7*cax(1) 1.1*cax(2)])
ylim([min(Freq) max(Freq)])
axis square;
pa_getpower(snd,Fs,'orientation','y'); % obtain from PandA
ylim([min(Freq) max(Freq)])
ax = axis;
xlim(0.6*ax([1 2]));
axis square;
box off;