![]() ![]() 11.25ĭesign an interpolator to change the sampling rate to 16 kHz with following specifications: Write a MATLAB program to implement the downsampling scheme, and plot the original signal and the downsampled signal versus the sample number, respectively. Hamming window required for FIR filter design (b) Generate a sinusoid with a 1000 Hz for 0.05 s using a sampling rate of 8 kHz, (a)ĭesign a decimator to change the sampling rate to 4 kHz with specifications below: Use MATLAB to solve Problems 11.24–11.30. This weighting is useful in implementation of the classifier, although theoretically not required. Finally, the cepstral features are weighted so that the range of all feature values is approximately equal. The cepstral domain representation has the advantage that the features are less correlated, which is important in efficient implementation of GMM-based classifiers. In the sixth step, the spectral-domain representation is translated into the cepstral domain cepstral features can be obtained by taking the inverse Fourier transform of the log-power spectrum. In the fifth step, the log of the energy in these frequency bands is computed. The frequency bands are spaced along the frequency axis according to the perceptually based nonlinear Mel scale, in which higher frequencies are represented with lower resolution. A filter-bank operation is applied to the power spectrum, thereby measuring the energy in different frequency bands. Third, the power spectrum is computed, without taking the log operation. Next, a Hamming window is applied to the frame a Hamming window reduces the effects of speech at the edges of the window, which is useful in obtaining a smooth spectral representation. First, the signal is pre-emphasized, which changes the tilt or slope of the spectrum to increase the energy of higher frequencies. Mel-frequency cepstral coefficient features are computed using a seven-step process. John-Paul Hosom, in Encyclopedia of Information Systems, 2003 V.B. Xlabel('Frequency (Hz)') ylabel('Amplitude | Y(f)| ') grid Y = filter(B,1,x) % perform digital filteringĪyk = 2⁎abs(fft(y))/N Ayk(1)=Ayk(1)/2 % single-side spectrum of the filtered data Wnc = 2⁎pi⁎900/fs % determine the normalized digital cutoff frequencyī = firwd(133,1,Wnc,0,4) % design FIR filter Xlabel('Frequency (Hz)') ylabel('Amplitude | X(f)| ') grid Xlabel('Number of samples') ylabel('Sample value') grid Īxk = 2⁎abs(fft(x))/N Axk(1)=Axk(1)/2 % calculate single side spectrum for x(n) V = sqrt(0.1)⁎randn(1,250) % Generate Gaussian random noise ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |