### XE31CZS exercise - Time-domain Noise Suppression Using Cumulative Sums.

• Signal preparation.
• Generate harmonic periodic signal, i.e. sinusoidal with following parameters:
- A = 1, f = 500 Hz, zero phase shift,
- fs = 8000 Hz, to = 0.1 s (duration).
• Load short speech signal (Czech word "reklama" SA106992.CS0, SA114992.CS0 - male voices, SA107992.CS0, SA110992.CS0 - female voices).
Format of these signals are binary. To read it into MATLAB environment use following attached function loadbin.m, sampling frequency of these signals is fs = 16 kHz .
Note.: Attention! This function is not standard function of MATLAB toolboxes! You should place it into your working directory!
• For both type of signals (sinusoidal and speech) create a mixture with zero mean white Gaussian noise. Signal-to-Noise Ration (SNR) of created mixtures should be: 10 dB, 0 dB, -10 dB.
Note.: for the mixture creation do not scale useful signal, scale just the additive noise by a constant k for given SNR (see lecture).

• Feed-back on generated signals.
• See always generated signals in time-domain.
• Listen the signals if it is reasonable (fcn sound).
• Evaluate back the SNR of the mixture.

• Time-domain Noise Suppression Using Cumulative Sums.
• Realize enhancement of useful signal from noisy background using cumulative sums, i.e. by addition of signals from 2 or more channels.
• Do not forget use different realization of white noise in particular channels.
• Evaluate SNR of output signal and compare the results with theoretically assumed SNR enhancement.