A classification method for environmental audio data

In area environment, variations of life form can exhale difference kind of audio signals, and these audio signals are nearly correlative with different biologic subsistence conditions and human activities. To analyze these audio signals automatically, in this paper, we propose a method, which employ effective segment length of audio data (ESLOAD), frequency component of maximum harmonic weight (FCOMHW) in the segment and first order difference Mel-frequency cepstral coefficients matrix (D-MFCCM) to classify area environmental audio data. In the method, it is used for short-time average magnitude to effectively segment the audio data, firstly; then FFT to calculate FCOMHW in the segment; finally, calculate D-MFCCM. For classifying operation, according to ESLOAD and FCOMHW, we confirm the searching range of every segment, individually confirm the segment possible audio type with corresponding D-MFCCM, and then confirm the possible audio types. Through the experiment of 9 categories total 107 normal area environmental audio data, it is indicated the method is effective to classify area environmental audio data.