Answers are grounded in this paper and cite their sources.
Abstract—Classification of common environmental background noise sources like train, airport, car and exhibition hall mixed with speech signals comprise of different utterances of various speakers required in many applications especially for forensic purposes. The acoustic features based on the Fourier and wavelet transformations are frequently used for environmental noise classification purpose. In the methodology presented, the noisy speech signal under test is initially decomposed into overlapped frames. In order to convert the latter into a set of bandlimited functions known as Intrinsic Mode Functions (IMFs), we use Empirical Mode Decomposition (EMD) method through Hilbert-Huang Transform (HHT). In this method the spectral and temporal features are extracted from the IMFs of different noisy speech signals and thereafter classification of different noise sources has been done using k-Nearest Neighbor (k-NN) classifier as a simpler way. Computed individual feature provides success rate of discrimination which varies from 77% to 85%. The combined feature vector enhances the success rate of classification as exhibited in the results. The technique presented reduces the dimensions of feature vector and proves the effectiveness of individual feature extracted from IMFs for the same.
No takes yet. Share an insight, caveat, or question.
Jhanwar et al. (2013) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: