Signalyze is an acoustic analysis program for the Macintosh platform. It can extract the features of speech that are of particular interest in this module. The following pages describe the basics of using the program. Help is available using the HELP option in the APPLE MENU (click on the Apple icon in the upper-left-hand corner of the screen). If you cannot solve a problem, alert the lab personnel. As a last resort, copies of the Signalyze manual are also available for examination in the lab room in case of problems.
Guides to the control panel are available from HELP and corresponding illustrations from the Signalyze manual are reproduced in the next few pages. Conceptually, everything you see in this program corresponds to the representations and measures covered by the Speech Production and Perception tutorial.
Check the input parameters
- in the MACRO menu, select RECORD SETUP
- select the APPLE SOUND INPUT DEVICE (fourth button down) for input
- set that the sampling frequency to 11025 Hz (that's the default, but you might need to adjust it)
- set the time limit to 3 seconds; a longer duration may be used if you wish, but this value should suffice for any one of the sentences
- save these settings by clicking SAVE
Start recording. Be prepared because when you start recording because it begins immediately. When you are recording, a box with a large microphone appears on the screen, and recording continues until the preset time limit (3 seconds in this case) has been reached; at that point the file is automatically displayed as a waveform in the uppermost of the 10 tracks on the screen.
- from the MACRO menu, select RECORD SIGNAL
Adjust the waveform display so that the entire length of the file fits within the display window. If the waveform is too big, make it smaller by using the ZOOM SIGNAL slider (the slider that is second from the top in the upper right side of the screen). It's easier to drag the slider box because clicking the arrows moves the waveform too quickly. You can listen to the recorded speech by selecting a section of the waveform with the mouse and clicking the speaker icon at the bottom right of the screen.Make a wideband spectrogram through the SPECTRAL menu. First, choose SPECTRAL ANALYSIS SET-UP, set the BAND value to WIDE and the range button to 0-5512 Hz (the FULL RANGE option). The other options can remain at their default settings. Now click the SPECTROGRAM-CEPSTROGRAM button at the bottom of the set-up box to draw a wideband spectrogram in the next available track. Once you have a wideband spectrogram, you can identify where the vowels are by their formants.
Extract the pitch of the utterance through the SPECTRAL menu. First, choose PITCH EXTRACTION SET-UP and set the FREQUENCY RANGE option by clicking on either the MALE or FEMALE buttons. All other values can remain at their default settings. Now choose DO. A contour should appear in the next open track, which traces the fundamental frequency of voiced segments of the utterance. [The actual value in Hertz at the cursor appears in the track information box to the left of the track, the lower number of the 2 in the upper left corner of the box]Measure the mean pitch of the segment from the pitch contour that was extracted. The data will appear in the window in the upper-left of the screen. First, choose the SHOW STATISTICS OPTIONS item from the EDIT menu. The values of interest here are the MAXIMUM and MINIMUM (left-hand column), and MEAN (right-hand column). De-select all other values. Now select the region of interest in pitch contour by setting a window around it. That window can include empty areas that correspond to unvoiced speech that has no associated pitch values. Having set the window, choose SHOW STATISTICS from the EDIT menu. Click DO. Three values appear in the data window, labeled MX (maximum), MN (minimum), and X (the mean).
Use linear predictive coding (LPC) to measure the frequency locations of the first two formants of voiced segments of interest in the utterance through the SPECTRAL menu. Choose SPECTRAL ANALYSIS SET-UP and then set the BAND value to LPC. You also need to set an appropriate number for order. Start with a setting of 14 and then adjust if necessary as described below. Now set the cursor in the time waveform to the center of the vowel of interest, and from the SPECTRAL menu, select SPECTRUM-CEPSTRUM.The point here is to create an accurate characterization of the formants in the vowel. Examine the peak locations shown in the LPC function, comparing them to the formants visible at the corresponding point in time in the wideband spectrogram. If the LPC is a good match, then the number of coefficients used is right. If one or more dark bands visible in the spectrogram are not visible as peaks in the LPC function, increase the number of coefficients by 2 and try again. Erase the existing function first (press the icon showing a WAVEFORM WITH A LINE THROUGH IT) and select that window as the destination for the new function (click on the ARROW icon that points into the window from the middle of the left-hand control panel). If peaks appear in the LPC function that do not reflect dark bands of energy on the spectrogram, decrease the number of coefficients by 2 and try again. Feel free to check your work with lab personnel.
Frequencies for the first two formants can be read by placing the cursor exactly at the highest points in each of the first two peaks in the LPC function. When the cursor is set, the corresponding frequency value appear as the upper number in the small data window in the control panel to the left of the function. This value is the frequency of the peak in Hz. (The number just below the frequency value is the amplitude, in dB. That value is not of immediate interest in this module.)