Noise pollution analysis and recognition
Moukas, P. (1982). Noise pollution analysis and recognition. (Unpublished Doctoral thesis, The City University)
Abstract
Instrumentation currently available for the automatic monitoring of noise nuisance has the shortcoming that although the intensity, duration and tie of occurrence of noises may be recorded, their source often cannot be identified.
This report describes research directed towards providing improved instrumentation which can identify noise pollution sources by exploiting the structure of the sounds they emit. The structure, which is due to the operating characteristics of the sources is revealed by the digital computation and display of Fourier spectrograms of digitized records of their sounds. In addition, the effect of ground reflections which introduces ripples in the spectra and is characteristic of the altitude of the source can be extracted by Cepstral analysis. Quasi-stationary fast periodicities of the signal appear in the spectrogram as quasi-straight lines along the time axis, whereas slow impulsive periodicities appear as isolated groups of spectra with increased total power and different structure.
Averaging spectra over a short periods (approximately 0.5 seconds) preserves the fast periodicities. The square root of the power in certain frequency bands of these average spectra can then be used as features in a feature space classifier. In a similar fashion, the cepstral “power” bands and the “energy” bands that result from the Fourier transform of the function representing the variation of the average amplitude with time can be used as additional features.
The scheme described above has been simulated on a general purpose digital computer and has been tested with real sounds of jet aircraft, helicopters and trains. A suboptimal search has been done to select the “best” feature ser, by sequentially designing and evaluating a linear classifier with one, two and so on features. Recognition better than 95% has been achieved using 6 spectral features only. When the additional features were included the performance was consistently improved by about 2% when 9 features were used. The effect of the width of the bands decreases with the number of features; however, 240Hz bandwidth was found optimum.
Publication Type: | Thesis (Doctoral) |
---|---|
Subjects: | Q Science Q Science > QA Mathematics > QA75 Electronic computers. Computer science |
Departments: | School of Science & Technology > Department of Computer Science School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses |
Download (247MB) | Preview
Export
Downloads
Downloads per month over past year