XE31CZS Exercise - Correlation Analysis and Applications
Tasks to do:
- Basic Properties of Autocorrelation Function
- Evaluate autocorrelation coefficients of given signal (fcn 'xcorr').
- Try options 'biased, unbiased, coeff, none' and compare
differences in results on following signals:
1. sinusoidal - s1 - f=15 Hz, fs=200 Hz, A=1, t=1 s ,
2. Gaussian white noise - b1 - power 0.7, mean value 0, fs=200 Hz, t=1 s,
3. Gaussian white noise - b2 - power 0.7, mean value 0, fs=200 Hz, t=10 s,
4. sinusoidal s1 + constant component 0.8,
5. noise b1 + constant component 0.8,
- Discuss different properties of given estimations.
- For signals defined in items 1., 2., 3. evaluate biased estimation
of coefficients R and R.
ATTENTION !!! To have comparable results, you must set always the
seed of random generator of Gaussian noise to 0 by using command
- Detection of the periodicity in signal using autocorrelation
- Estimate fundamental period of voiced sounds vf5.bin and vm5.bin
on the basis of second main maximum of autocorrelation function
- Discuss the problems of this approach.
- Try to estimate the delay between signals from two different input
channles using cross-correlation function.
Sampling frequency of given signals is 8 kHz.
Clean speech signals: s0001-l.bin a s0001-r.bin
Speech with noisy background: x0002-l.bin a x0002-r.bin
- a) evaluate cross-correlation from whole signal
- a) evaluate cross-correlation from segmented signal; frame length
wlen = 256, 512 samples without overlapping.