This document commonly with all following links uses ISO
8859-2 (ISO-Latin-2) encoding
AE2M31RAT, AE2M31ZRE - seminar - TASK No. 2
Basic time-domain and spectral characteristics
Tasks to do:
- Computation of basic time-domain characteristics
- Create MATLAB functions for the computation of basic time domain
characteristics in particular short-time frames. Inputs
for these functions should be:
- vector with analyzed signal
- sampling frequency of the signal
- length of short-time frame in miliseconds
- step (time movement) of short-time frame in miliseconds
- Output parameters will be:
- 1st output parameter - column vector (matrix) with estimated
characteristics, where each row will contain characteristics for one
analyzed short-time frame,
- 2nd output parameter - column vector containing the time of each frame
- Use the following signal for the first
attempt SA176S01.CS0 - raw
data, fs=16000 Hz, for loading into MATLAB
use function loadbin.m
- Possible structure for the computation of particular
characteristics - vykon.m
- Compute following parameters:
- Power in dB
- Peak-to-peak value
- Zero-crossing-rate (function 'signum' is avaialble in MATLAB as sign)
- Starting time of each short-time frame
- 1st checked result : time-dependency of
signal short-time energy for
SA176S01.CS0. Compute it for legnths
30ms, 10ms, 5ms, 1ms.
- Repeat for the signal mc20bc116016.ils_a - raw
data, fs=44100 Hz.
- 2nd checked result:waveform and
time-dependency of energy in dBs in on-line recorded signal for the
short-time frame length 32ms with 50% overlapping.
- Computation of spectral charactersitics
- Compute spectrogram for above mentioned signals with the
following setup of spectral analysis (observe and explain differences):
- short-time frame length 32 ms, frame overlapping 50%, Hamming window
- short-time frame length 5 ms, frame overlapping 50%, Hamming window
- short-time frame length 5 ms, frame overlapping 50%, Hamming
window, zero padding to NDFT 512
- For above mentioned examples observe also short-time spectrum for
selected frame (voiced and unvoiced).
- Observe the influence of preemphasis in speech spectrogram.
- 3rd checked result: wavefrom and
spectrogram of signal SA176S01.CS0 for the
short-time frame length 32ms with 50% overlapping WITHOUT preemphasis
- 4th checked result: wavefrom and
spectrogram of signal
SA176S01.CS0 for the
short-time frame length 32ms with 50% overlapping WITH preemphasis
- Observe waveforms and
spectrograms also for on-line recorded signals.
Further tasks for a homework:
- Comparison of power analysis in MATLAB, Praat, and Wavesurfer
- Work with the
- raw data without header, fs=44100 Hz,
- observe power in Praat (menu Intensity),
- compute the power in MATLAB with the setup equivalent to Praat and
- observe power computation also within Wavesurfer (panel Power plot) and
set the parameters to obtain same results as in the preceeding two
cases in Praat and MATLAB.
- Comparison of spectral analysis in MATLAB, Praat, and Wavesurfer
- For signals SA176S01.CS0 (raw data, fs=16000 Hz) and mc20bc116016.ils_a
(raw data, fs=44100 Hz) observe spectrograms in
Praat and Wavesurfer.
- Observe spectrograms for various setups, compare mainly the
differences for varying frame lengths and preemphasis setup.