Ceska verze teto stranky

Back to main page | list of seminars

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 beginning.
  • 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 - mypwr.m
  • Compute following parameters:
    • Energy
    • Power
    • Power in dB
    • RMS
    • Intensity
    • Peak-to-peak value
    • Zero-crossing-rate (function 'signum' is avaialble in MATLAB as sign)
    • Starting time of each short-time frame
  • 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.
  • Result:
    - Display waveform and time-dependency of signal power and power in dBs in on-line recorded utterance of the legnth 5s for the short-time frame length 32ms with 50% overlapping. Signal parameters: sampling frequency 16 kHz, 16 bit linear PCM, 1 channel (mono signal).
    - Estimate SNR from observed short-time power in dBs.

• 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.
  • Result:
    - Observe wavefrom and spectrogram of for an on-line recorded utterance of the length 5 s using short-time frame length 32ms, 50% overlapping, and WITHOUT and WITH preemphasis application.

• Comparison of power analysis in MATLAB, Praat, and Wavesurfer
  • Work with the signal mc20bc116016.ils_a - 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 compare results,
  • 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.