MEA - Spectrum Measurement Program

I. General Description

This program displays spectra, one at a time, from a waveform file. It is designed to step through the file or move randomly about the file to allow measurement of speech parameters. Each spectrum is displayed with estimated envelope and formant frequencies. Most variations in the way that the spectrum is computed are under program control and can be set with commands. This allows considerable variation in the way that the program can be used. Additionally, the program has the capability of opening an auxiliary log file. When a log file is open, formant frequency estimates and associated amplitudes can be written to the file for any frame that is being viewed using the SAVE command. The data for each frame is written to the file as a list of numbers on one line. The first value is the time offset of the frame given in msec from the start of the file. The remaining numbers are in pairs with the first number being a frequency and the next number an amplitude at that frequency. The first 5 such pairs pertain to the first 5 formants (F1 - F5) respectively. The final two pairs are an additional pole and spectral zero, as for example might be present with a nasal sound. All of these numbers are simply the frequency estimates the program has demarked in the display with arrows. The program guesses the right values for each, but it is your responsibility to see that they are actually set correctly. Specific commands that can be given, subcommands, and comments about what they all mean follow.

II. Command Summary

OPEN -- Open a log or waveform file

Syntax
OPEN file_type file_name
Description
Opens log or waveform files for use by MEA. "file_type" may be either WAV for waveform or LOG for logfile. The file_name is the name of the logfile or the waveform file to be opened. If the named log file does not exist, it will be created. If the log file already exists, it will be opened and new frames of data will be appended to the end of the previously saved frames of data.

Examples

OPEN WAV \PATH\SPEECH.WAV 
OPEN LOG STOPCONS.MEA

SAVE -- Save current data to logfile.

Syntax
SAVE
Description
Writes the current amplitude and frequency marks to an open log file. In the current PC version of MEA, the SAVE command must be used explicitly for each analysis frame that is to appear in the log file.

CLOSE -- Close the log file

Syntax
CLOSE
Description
Closes the currently open Log file. Opening a new log file and exiting the program also close any open log file. Note that files must be closed to be saved. This command allows you to stop writing information into the log file while continuing to view the waveform.

EXIT -- Exit the MEA program

Syntax
EXIT
Description
Exit from measure to the operating system level.

<time> -- Position display to specified time

Syntax
time_in_msec
Description
Simply entering the time in msec (as offset from the start of the waveform) causes MEA to jump to that point in the file and display the spectrum computed from that time frame. Note that the actual time that MEA positions itself to will always be an integer multiple of the STEP size.
Examples
200<Enter> [move to frame 200 msec from start of file]

<return> -- STEP to next analysis frame

Syntax
<return>
Description
Pressing return or enter will cause MEA to advance to the next analysis frame (as specified by the STEP size) and to display the spectrum computed for that frame.

HELP -- Show help screen

Syntax
HELP
Description
Displays a screen that contains a summary of the commands, parameters and options that the MEA program recognizes. On the PC version, the program looks for the file \DOC\MEA.HLP and prints that to the screen.

SET -- Set various program options

Syntax
SET mode
SET parameter value
SET parameter subparameter value
Description This command allows various analysis parameters to be set or changed. As the syntax implies, there are three basic forms of the set command. One simply sets a mode of operation, the second sets a value for some parameter, and the third sets a value for a subparameter. The following describes each alternative:
- Modes
  - HIPASS
    Apply high frequency preemphasis before computing the spectrum. This reduces low frequency energy and tilts the spectrum up by 6 dB / Octave.
  - NOHIPASS
    Disables the HIPASS option.
- Parameters
  - STEP
    Step size for advancing through the file (in msec).
  - SMOOTH
    Determines which smoothing algorithm (AVG, LPC, or CEP) is used to estimate the smoothed spectrum. By default, MEA simply smooths the raw DFT data with a lowpass binomial filter function to obtain a weighted running average. This option is called AVG. The LPC algorithm is a covariance method, all-pole model. CEP stands for Cepstrum and is another way to remove spectral periodicity due to the fundamental frequency.
  - NPASS
    This parameter always relates to the smoothing algorithm used for obtaining the spectral envelope (See SET SMOOTH). When smoothing is obtained by a simple running average (AVG), NPASS is the number of smoothing passes in computing the envelope of the spectrum. The more passes, the smoother the envelope. When smoothing is obtained via LPC analysis (LPC), NPASS is actually the number of terms in the inverse filter polynomial. In this case, smaller values of NPASS lead to greater smoothness (fewer peaks) in the smooth spectrum. For the cepstrum (CEP) smoothing, NPASS is the number of cepstrum coefecients to retain. Small numbers produce more smoothness.
  - NFFT
    Number of transform points in the discrete Fourier transform that is used to compute the spectrum. Normally, this is set by the program based on the window length. If you wish to change it, you must use a value larger than the one the program chose.
  - WINDOW
    The analysis window is the chunk of waveform that is actually analyzed in computing the spectrum. This is different than the step parameter. The window length is usually longer than the step size. This means that analysis windows often overlap from one step to the next. Longer analysis windows provide better frequency resolution, but, if the speech signal is changing long windows can average over important changes. So there is a trade off between frequency resolution and temporal resolution. MEA lets you adjust the window length to see the effects of rapid changes (say in formant transitions) with short windows and still use long windows for analysis of vowels where the spectrum is not changing rapidly. Window length is specified via the LENGTH subcommand.
    Another important characteristic of the analysis window is its 'shape'. Shape means how the speech samples are weighted at different points in the window. For instance, a window that weights all samples the same whether they are in the center of the window or toward its edges is called a RECTANGULAR window. Various kinds of windows have been designed which weight the samples in the center of the window more than ones at the edges. These windows are named after people responsible for studying their effects on signal analysis. Two such windows are the HAMMING window and the HANNING window. The HAMMING window tapers off to zero at the edges so that speech samples at the edge of the window are always weighted by zero (i.e., they really don't effect the analysis). The HANNING window is a little less severe: the edges of the window taper off to almost zero. The choice of window type, especially between rectangular and the tapered kinds, can have a large effect on the analysis. The reasons for this effect are too technical to go into here, but have to do with the analysis trying to account for discontinuities at the edge of the window. Unless you know what you are doing, it is best to stick with the default window type used by MEA. The window type is given via the TYPE subparameter.

Examples

		SET HIPASS
		SET NFFT 512 
		SET STEP 5 
		SET SMOOTH LPC
		SET NPASS 15
		SET WINDOW TYPE HAMMING 
		SET WINDOW LENGTH 25.6

MARK -- Mark formants

Syntax
MARK
Description
MARK places the program in a special mode of operation that enables you to adjust the locations of the arrows demarking formants and/or zeroes in the spectrum. This mode can also be used for simply measuring things of interest without changing mark locations. The program tries to estimate formant frequencies using a combination of information from the current frame and from the previous frame. This process is prone to error, and hence formant estimates may need to be corrected if you are logging the information. To do so, issue the mark command, position the cursors (using the arrow keys) to the frequency-amplitude coordinates that are correct then press the key designated in the MARK screen menu. E.g., press a '1' for F1; a '2' for F2; a 'P' for extra POLE; a 'Q' to QUIT marking, etc. While you are adjusting the mark locations extra marks will appear on the screen (where you indicate formants should be) without erasing the old formant marks. When you QUIT marking, the display will be updated to erase the old marks and show only the new ones; until then all marks remain on the screen.