A highly robust audio fingerprinting scheme in real environments

Abstract

Recently, content-based audio identification techniques by an audio fingerprinting scheme that can retrieve audio information without any text-based query. They have been recognized as one of the state-of-the-art and attractive application services on the music portal market in wire/wireless communications. This dissertation introduces a methodology to this challenging task using an audio signal query to retrieve polyphonic music items by matching it to pre-indexed audio references. In real environments, however, sound recordings are commonly distorted by channel and background noise. As well, music signals can be easily distorted by time stretch (tempo change). The performance of audio identification is greatly degraded by those distortion factors. Thus, the robustness of an audio fingerprinting system is still one of the most important issues in music information retrieval by content-based audio identification techniques. This dissertation introduces the conventional audio fingerprinting schemes such as stochastic modeling and audio hashing. In the stochastic modeling scheme, spectral parameters are conventionally used to build a stochastic model. Foote proposed the stochastic modeling method for content-based music information retrieval (MIR) [16]. The stochastic model is based on the spectral envelope histogram, the histogram of spectral audio feature counts at the code vectors of vector quantization (VQ). In this dissertation, we propose a new distance metric to measure the similarity of two probability distributions and apply the dynamic matching method instead of the static matching method. As well, we proposed the stochastic modeling method which uses pitch histogram instead of spectral envelope histogram. Music can be identified by distinctive melody lines. Melody line consists of the harmony of musical notes. After all, pitch becomes very useful information because it is a basis of melody note. In addition, the number of histogram bins can be limit...