Towards the Application of Fuzzy Logic to the Sound Recordings Fingerprint Computation and Comparison LADISLAVA JANKU, LENKA LHOTSKA Department of Cybernetics, Faculty of Electrical Engineering Czech Technical University in Prague Technicka 2, 166 27 Praha 6 CZECH REPUBLIC Abstract: - This paper deals with the application of computational auditory scene analysis system and voice identification algorithms in audio fingerprinting. Generally, audio fingerprinting technology generates a unique fingerprint for an audio file based on an analysis of the acoustic properties of the audio itself and represent unique identification key of the music piece. Presented approach enables extract voice features and compute the hash function, which depends on these voice properties. A system for voice extraction through exploration of the auditory scene analysis approaches and fuzzy rule system is presented. Key-Words: - fuzzy logic, computational auditory scene analysis, audio fingerprint 1 Introduction This study is motivated by an interest in the properties of sound signals and identification of them with the practical application of automatic searching in large music databases containing music data in signal representation or in compressed format. In recent years, audio fingerprinting technology was suggested as a sufficient and powerful tool for audio sample identification. Audio fingerprint is a unique key of each sound sample. This property brings about the following effect – a music piece, performed by one performer, is characterized by the different audio fingerprints that the same music piece performed by another musician. This property of audio fingerprinting causes a limitation of searching facilities. It’s not possible to use them for melody-oriented or performer- oriented search. This weakness can be partially avoided by extension of the audio fingerprint generation algorithm. Approach presented in this paper involves physiologically motivated sound pre-processing, voice feature extraction and hashing. It combines model of auditory periphery with audio features extracting mechanism, and fuzzy rule system. 2 Audio Fingerprints - Background Audio fingerprinting technology generates a unique fingerprint for an audio file based on an analysis of the acoustic properties of the audio itself. Each audio fingerprint is unique, e.g each sound has a unique audiofingerprint. Audio fingerprint is generated by the application both of acoustical features extracting algorithms and by one-way hash functions. Figure 1 shows the whole process of the audio fingerprint generation. Input sample is divided into 30-70 narrow frequency bands. Fig 1.: Standard audio fingerprint generation method Signal characteristics or audio features are extracted from each frequency band; these features are measured and used as a background for the fingerprint