Font Size: a A A

Audio Perceptual Hash Algorithm Based On Auditory Filter And Its Application In Music Information Retrieval

Posted on:2016-08-28Degree:MasterType:Thesis
Country:ChinaCandidate:J H MengFull Text:PDF
GTID:2298330467977356Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
With the development of multimedia and Internet technologies, more and more digital audio resources can be acquired easily, human being have excellent ability to identify different music just in a few seconds, even in noisy environment. The question is that how to classify and manage huge number of audio resources automatically by computer. That leads to the audio perceptual hash algorithm, a scheme used for content-based music information retrieval.Given the current shortcomings in many papers, such as robustness and complexity, we propose a new audio perceptual hash algotithm. First, a novel time-frequency representation for audio features is proposed. Audio signal was filered by a multiband Gammchirp filterbank within sensitive frequency band. After that, calculate energy spectrum of each band of each frame. It is shown to achieve good performance in robustness and resisting geometric distortion. Then, the local feature of the Gammachirp energy spectrum is extracted by non-negative matrix factorization (NMF). Finally, audio perceptual hash is generated by applying difference and quantization on the extracted features. Experimental results illustrate that the proposed algorithm achieves superior performance in identification rate both in the software simulating environment and practical environment.On the other hand, retrieval speed is also an important problem for automatic recognition of music. Simply modify the algorithm has been unable to obtain significant performance improvement. So, it is necessary to use other computing devices to improve the audio retrieval algorithm retrieval speed. Graphic Processing Unit (GPU) can provide a powerful floating-point computing power, use of GPU accelerating existing audio retrieval algorithm is of great significance. In this paper, with the cooperation of CPU and GPU, not only the match time consuming, but also the overall retrieval time consuming are drastically reduced.Finally, we designed a music information retrieval system based on the proposed scheme. It can identify an unknown music by recording just a few seconds.
Keywords/Search Tags:Audio perceptual hash, Gammachirp filter bank, Non-negative matrixfactorization, GPU parallel computing
PDF Full Text Request
Related items