Abstract—Classification of common environmentalbackground noise sources like train, airport, car andexhibition hall mixed with speech signals comprise ofdifferent utterances of various speakers required in manyapplications especially for forensic purposes. The acousticfeatures based on the Fourier and wavelet transformationsare frequently used for environmental noise classificationpurpose. In the methodology presented, the noisy speechsignal under test is initially decomposed into overlappedframes. In order to convert the latter into a set of bandlimitedfunctions known as Intrinsic Mode Functions (IMFs),we use Empirical Mode Decomposition (EMD) methodthrough Hilbert-Huang Transform (HHT). In this methodthe spectral and temporal features are extracted from theIMFs of different noisy speech signals and thereafterclassification of different noise sources has been done usingk-Nearest Neighbor (k-NN) classifier as a simpler way.Computed individual feature provides success rate ofdiscrimination which varies from 77% to 85%. Thecombined feature vector enhances the success rate ofclassification as exhibited in the results. The techniquepresented reduces the dimensions of feature vector andproves the effectiveness of individual feature extracted fromIMFs for the same.

Index Terms—Hilbert-huang transform, intrinsic modefunction, empirical mode decomposition, k-nearest neighborclassifier

Cite: Deepak Jhanwar, Kamlesh K. Sharma, and S. G. Modani, "Classification of Environmental Background Noise Sources Using Hilbert-Huang Transform," International Journal of Signal Processing Systems,  Vol. 1, No. 1, pp. 68-73,  June 2013. doi: 10.12720/ijsps.1.1.68-73