XE31CZS Exercise - Linear Prediction Analysis (LPC) - LPC Spectrum
Tasks to do:
- Compare short-time spectrum estimation using DFT and LPC analysis
- For the evaluation of algorithm use voiced part of speech vm0.bin.
- AR cofficients of the signal could be computed using fcn
lpc. Parameters of AR model will be p=16, N = 512
- LPC spectrum is given as frequency response synthesizing filter
( fcn freqz )
- Draw both short-time spectrum estimations (DFT and LPC) into one
plot for comparison !!
- Observe influence of wieghting of analyzed signal by Hamming window.
- Observe influence of LPC spectrum on amplification parameter G.
- Compare G or E_p respectively evaluated as:
- output of fcn lpc,
- by the formula from AR and autocorrelation coefficients,
- by the direct evaluation of power of error signal.
- Observe the results for different order of AR model p=16, 10,
6. Explain the results !!
- Example: Compare possible spectral resolution in spectral
analysis using DFT aan LPC.
- Choose optimal order of AR model for particular task specified bellow.
- Evaluate spectral estimation from follwoing number of samples: N=64 and N= 256.
- Use weigthing of the signal by Hamming window.
- sinusoidal: f=13 Hz, fs= 100 Hz
- 2 sinusoidals of close frequencies: f1=13 Hz, f2=15 Hz, fs= 100 Hz