Speech Processing
by Computer
FILTERBANK
ANALYSIS-SYNTHESIS
This lab session
demonstrates how a rough spectral analysis of a signal may be obtained through
the use of a bank of filters. The
combination of the spectral analysis with some information about the excitation
allows the recreation of the signal with some interesting modifications.
1. Low-pass/High-pass
and Band-pass filtering
(i) Acquire a speech signal at 20000
samples/second.
(ii) Low-pass filter the signal at 1000Hz and
replay the result.
(iii) High-pass filter the original signal at
900Hz and replay the result.
(iv) Band-pass filter the original signal
between 900 and 1000Hz and replay the result.
(v) Compare spectrograms of the four signals.
2. Filterbank
analysis
(i) Acquire a speech signal and a
laryngograph signal at 20000 samples/second.
(ii) Use the ‘19-channel auditory filterbank’
program to save the energies in 19 frequency bands from the speech signal.
(iii) Use the ‘make grey scale’ option to turn
the energies into a grey-level picture.
Dsiplay the grey-level picture of the filterbank outputs alongside a
spectrogram of the original signal.
3. Filterbank
synthesis
(i) Convert the laryngograph waveform to Tx
pulses.
(ii) Convert the Tx data into a fundamental
frequency contour.
(iii) Use the '19 channel synthesizer' program
to resynthesize the original signal from the filterbank analysis and the
fundamental frequency contour.
(iv) Use the '19 channel synthesizer' program with
the 'Whispered’ switch to produce a whispered version of the utterance. What excitation is being used here?
4. Synthesis
with alternative excitation.
(i) Acquire into the same file as the
original speech, a piece of orchestral music at 10,000 samples/sec.
(ii) Using the '19 channel synthesizer'
program with the 'alternative excitation' switch (and the music item number),
resynthesize the speech signal using an orchestral excitation. Replay the result. Compare spectrograms between the different resynthesized versions
and the original signal.