Oliver Hellmuth

Robust matching of audio signals using spectral flatness features

Stimulated by the ever-increasing amount of available multimedia data, content-related techniques for the management of audio material have received much interest recently. This paper discusses the problem of robust identification of audio signals by matching them to a known reference. In order to perform well under realworld conditions, the matching process needs to rely on features which are robust with respect to common signal distortions. A family of suitable features with favorable properties is proposed and evaluated for their recognition performance. Applications of signal matching, including fingerprinting, are discussed.

Scalable robust audio fingerprinting using MPEG-7 content description

Much interest has recently been received by systems for audio fingerprinting which enable automatic content-based identification by extracting unique signatures from the signal. Among other aspects, the main requirements for such systems include robustness to a wide range of signal distortions and availability of fast search methods, even for large fingerprint databases. This paper describes the provisions of the MPEG-7 standard for audio fingerprinting which allow for interoperability of fingerprints generated according to the open standardized specification for extraction. In addition, it discusses the ability to generate scalable fingerprints providing different trade-offs between fingerprint compactness, temporal coverage and robustness of recognition, and gives experimental results for a number of system configurations.

Robust identification/fingerprinting of audio signals using spectral flatness features

Recently, the problem of content‐based identification material has received increased attention as an important technique for managing the ever‐increasing amount of multimedia assets available to users today. This talk discusses the problem of robust identification of audio signals by comparing them to a known reference (‘‘fingerprint’’) in the feature domain. Desirable properties of the underlying features include robustness with respect to common signal distortions and compactness of representation. A family of suitable features with favorable properties is described and evaluated for their recognition performance. Some applications of signal identification are discussed, including MPEG‐7 Audio.