The Dynamic Sub Nyquist Sampling Model For Compressive Sampling Based Music Information Retrieval

Download or read book The Dynamic Sub-nyquist Sampling Model for Compressive Sampling Based Music Information Retrieval - written by and published by . This book was released on 2012 with total page 478 pages. Available in PDF, EPUB and Kindle. Book excerpt: Audio identification is a task that can be accomplished through the use of various methods, including tagging, encryption, watermarking and so on. Fingerprinting however, is one method that has several advantages over the other approaches. Audio fingerprinting systems consist of two phases: the preparation phase and the recognition phase. The first phase involves building a database with the audio fmgerprints of many songs and their associated metadata. In the second phase, the fingerprint of an unknown song is extracted and compared with those in the database. What is the best method for extracting data from a piece of unknown audio? Compressive sensing (CS) (also referred to as compressive sampling) has attracted increasing interest from a wide range of researchers in various fields including signal processing, image processing, information theory, mathematics, computer vision, pattern recognition and so on. The basic principle of CS is that it exploits signal sparsity to reduce the number of measurements needed for digital acquisition, thus enabling low-rate sampling and high-resolution sensing. In other words, CS theory provides a possible way of recovering sparse signals by projecting them onto a small number of random vectors. This dissertation proposes the use of compressive sensing (or compressive sampling) for audio extraction, based on a compact and robust audio fmgerprint system. We begin by investigating the fundamental aspects of building such a system. Next, we construct a practical implementation of an analogue signal-to-information converter to acquire the sub-Nyquist sampling data. Thirdly, we identify efficient features, which are invariant to time and frequency distortions, and test different data search algorithms to optimise search efficiency. Finally, we conduct various simulations to establish the feasibility and efficiency of the proposed approach.