Combining multi-probe histogram and order-statistics based LSH for scalable audio content retrieval

Abstract

In order to improve the reliability and the scalability of content-based retrieval of variant audio tracks from large music databases, we suggest a new multi-stage LSH scheme that consists in (i) extracting compact but accurate representations from audio tracks by exploiting the LSH idea to summarize audio tracks, and (ii) adequately organizing the resulting representations in LSH tables, retaining almost the same accuracy as an exact kNN retrieval. In the first stage, we use major bins of successive chroma features to calculate a multi-probe histogram (MPH) that is concise but retains the information about local temporal correlations. In the second stage, based on the order statistics (OS) of the MPH, we propose a new LSH scheme, OS-LSH, to organize and probe the histograms. The representation and organization of the audio tracks are storage efficient and support robust and scalable retrieval. Extensive experiments over a large dataset with 30,000 real audio tracks confirm the effectiveness and efficiency of the proposed scheme.

Adaptive Temporal Modeling of Audio Features in the Context of Music Structure Segmentation

“…The MPH is a fixed-size representation of a chroma sequence which is determined by the dominant pitch classes transition between all adjacent frames of the sequence. A more detailed description can be found in [12]. Such histograms reflect the tonal structure of local portions of the audio and allow to model the evolution of the tonal context through time and capture slow varying harmonic patterns that relate to musical patterns.…”

“…In [8] a contextual measure of similarity that considers sequences of feature frames instead of single frames in the similarity matrix computation allows to strengthen the visualization of repetitive patterns in the audio features. In [6], the evolution of the tonal context in the audio signal is described by concatenating mid-term chroma sequences in Multi-Probe Histograms [12].…”

Recognition and Summarization of Chord Progressions and Their Application to Music Information Retrieval

“…(iv) Global summarization retaining harmonic progressions. A multi-probe histogram [5] is calculated from the chroma feature sequences by heuristically probing the transition between major bins of adjacent chroma features, which is able to retain local spectral and temporal information to some degree. It is more concise compared with local summarization, and is more accurate than conventional global summarization by retaining the temporal information of music signals.…”

“…for t = 2, 3, · · · , l do Forward iteration 5: for t = l − 1, l − 2, · · · , 1 do Reverse iteration 11: S t ← j∈y t+1 p t+1,j 12:…”

“…Some significant research progress has been made during the past few years. Music signals usually are described by sequences of low-level features such as short-time Fourier transform (STFT) [1], pitch [2], Melfrequency cepstral coefficient (MFCC) [3], and chroma [4], [5]. Unfortunately, among most existing work, music audio content analysis and summarizations, by means of these low-level features, are inefficient and inflexible for a scalable music information retrieval (MIR) task since music knowledge is seldom considered.…”