An approach to calculate a feature vector using Fourier and Constant-Q transforms for audio fingerprinting

Mansurov Alexandr Valerievich, Ladygin Pavel Sergeevich

This paper presents an approach for audio fingerprinting of audio files based on the calculation of a feature vector using Fourier and Constant-Q transforms. A comparative analysis of spectrograms and chromatograms obtained with the Fourier and Constant-Q transforms is performed, and peculiarities of the spectrograms and chromatograms are discussed. It is concluded that the calculation of feature vectors by processing chromatograms to identify roots of chords (or “pure” tones) is more effective than using techniques that require processing spectrograms. The proposed approach for the calculation of feature vectors allows their simple comparison and identification of complex musical compositions with an accuracy of 97%.

Keywords:audio fingerprinting, audio files, spectrogram, identification, chromatogram, wavelet, constant-q transform