LPC Spectrum Analyzer Software

In 1967, Burg reported on a new spectrum analysis method called MEM（Maximum Entropy Method）for analysis of seismic waves. Because the MEM is able to obtain a high resolution spectrum from transient signals including seismic waves, MEM application has spread to the study of geomagnetic variations, solar cycles and voice recognition since the 1970s. It also goes by other names including AR（auto-regression）and LPC（Linear Predictive Coefficient）in these studies. The MEM, however, has not spread to other general fields due to the heavy calculation load that prevents realtime processing. However, the recent development of high-speed microprocessors has achieved LPC spectrum real-time processing, enabling the LPC spectrum analysis to be applied to many fields.

We provide the spectrum analyzer software that displays LPC spectrum of various sounds including bird call, voice, impact sound, seismic wave and long steady-state signal. The software works on Microsoft Windows. First, you record the sounds into the WAV files using a data recorder. As shown in Figure 4, the software gives FIR digital filter processing to the waveform of sound (WAV file). Then, the software calculates “One-dimensional LPC spectrum”, “Two-dimensional LPC spectrum” and “Sonogram”.

First, in the three RadioButtons at the bottom left of the menu screen shown in Figure 5, you click to select one from among ●Bird call or Voice (Transient signals) ●Concrete (Impact sounds) ●Machine (Steady-state signals). Then, the software segments waveforms of the target sound automatically from a continuous recording. Next, in the six Buttons at the top of the menu screen, you click to select one from among “LPC 1-d”, “LPC 2-d”, “Sonogram View” and other things. Then, the software displays “One-dimensional LPC spectrum”, “Two-dimensional LPC spectrum”, “Sonogram” and other things.

At the top of the screen shown in Figure 6, the software displays the oscillogram and the one-dimensional LPC spectrum of the WAV file that is selected from the first cluster. Similarly, in the middle of the screen, the software displays the oscillogram and the one-dimensional LPC spectrum of the WAV file that is selected from the second cluster. Note that, the software displays the filtered oscillograms by the FIR digital filter, and extracts the one-dimensional LPC spectra from the cyan-colored waveforms, respectively. We can compare the one-dimensional LPC spectrum shown at the top of the screen with the one-dimensional LPC spectrum shown in the middle of the screen. Furthermore, at the bottom of the screen, the software displays the list of the peak frequencies in the one-dimensional LPC spectra of all WAV files.

On the left of the screen shown in Figure 7, the software displays the oscillogram and the two-dimensional LPC spectrum of the WAV file that is selected from the first cluster. Similarly, on the right of the screen, the software displays the oscillogram and the two-dimensional LPC spectrum of the WAV file that is selected from the second cluster. Note that, the software displays the filtered oscillograms by the FIR digital filter, and extracts the two-dimensional LPC spectra from the cyan-colored waveforms, respectively. We can compare the two-dimensional LPC spectrum shown on the left of the screen with the two-dimensional LPC spectrum shown on the right of the screen. Furthermore, at the bottom of the screen, the software displays the list of the peak frequencies in the two-dimensional LPC spectra of all WAV files.

In the middle of the screen shown in Figure 8, the software displays the oscillogram and the sonogram of the WAV file that is selected from the cluster. Note that, the software displays the filtered oscillogram by the FIR digital filter, and extracts the sonogram from the cyan-colored waveform. The software has an enlargement display function, a scroll display function, a monochrome display function, and other functions. When the vertical red line is set in the middle of the screen using a computer mouse, on the right of the screen the software displays the cross-section of the sonogram in the position of the vertical red line as a one-dimensional LPC spectrum. Furthermore, in the window shown in Figure 9, the software displays the FPE (Final Prediction Error) and the AIC (Akaike Information Criterion) of the one-dimensional LPC spectrum.

On the screen shown in Figure 10, the software displays the result of the segmentation that segments waveforms of the target sound automatically from a continuous recording. We can confirm the result of the segmentation for the one-dimensional LPC spectrum, the two-dimensional LPC spectrum and the sonogram. In order to reduce the processing overhead, the recognition software does not display the image shown in Figure 10. We can confirm the result of segmentation using the segmentation software. Figure 11 shows the result of segmentation in the case where the target sounds are close. In order to distinguish the target sounds that are close, the software displays the target sounds using two colors (magenta and cyan) alternately. The software extracts LPC spectra from the magenta waveforms and the cyan waveforms respectively.

This is a simulation software for the FIR (Finite Impulse Response) digital filter. At the top right of the screen shown in Figure 12, we set “type of filter” and “characteristic values of filter”. At the bottom right of the screen, we set “sampling frequency” and “frequencies of two input waves”. These two input waves are added and input into the FIR digital filter. Figure 13 typically shows the relationship between the input wave (green) and the output wave (yellow) in FIR digital filter. At the top of the screen shown in Figure 12, the input waveform (green) and the output waveform (yellow) are displayed to overlap each other. Moreover, at the bottom of the screen shown in Figure 12, the linear transfer function or the logarithmic transfer function is displayed.

On the screen shown in Figure 14, we set the values for “FIR Digital Filter”, “Waveform”, “Segmentation”, “Linear Predictive Coefficient”, “Geometric Distance” and “Result”. We can “Save these values As” and “Open them”. Furthermore, we can “Save them As” and “Open them” in “LPC 1-d”, “LPC 2-d” and “Sonogram View”.