Tempo Automatic Detection In MP3 Compressed Domain

Tempo Automatic Detection in MP3 Compressed DomainThe research of tempo automatic detection is composed of the study of musical theoretics,signal processing and pattern recognition.And it is significant for the management of the music digital library,thc music search in the internet and the daily amusement.After studying the musical theoretics,the musical conceptions,the methods to describe the musical features and the theoretics of musical signals analyzing,some signal processing conceptions and algorithms and signal features about music tempo detecting is prcscntcd.Music is composed by a scrics of notes in sequence,each note also contains three parts of characteristics:pitch,duration, intensity.Musical signal analysis, including musical signal pre-processing and related characteristics. The characteristics of musical signals are Short-time Average Energy,Zero-crossing Rate,Linear Prediction Coefficients,Power Spectrum,MF CC Coefficients and Entropy.In the second part of the article,I briefly introduced three basic methods of tempo detection.firstly,a tempo detection via the periodicity transform.Key portions of the method are the feature vectors,a new method of resampling(a step required by the periodicity transform) ,and a modification to the periodicity transform that allows it to consider only periodicities within some prespecified range.Secondly,using an autocorrelation phase matrix and Shannon entropy to detect tempo,this algorithm is based on autocorrelation.What distinguishes it from other autocorrelation approaches is that we computes the distribution of lag autocorrelation energy as a function of phase.This results in a lag-by-phase matrix that compactly represents the repetitive structure of a musical signal.In some signals where autocorrelation does not work,to compute tempo,we enhance a standard autocorrelation of the signal with the Shannon entropy of the phase information in the matrix.The formula for the lag k cross-correlation C_k between signals x₁ and x₂ is:Autocorrelation is a special case of cross-correlation where x₁=x₂.Namely the autocorrelation A of a signal X is:where f ft is the fast Fourier transform,if ft is the inverse fast Fourier trans-form.Shannon entropy H:Lastly,using beat histograms to detect tempo.The algorithm is mostly based on self-similarity rather than onset detection.However,an onset detection component is used to calculate the phase of the dominant periodicities.Multiple frequency bands are calculated using a Discrete Wavelet Transform.Subsequently the envelope of each band is extracted and autocorrelation is used to find the dominant periodicities of the audio signal.These dominant periodicities are accumulated into a Beat Histogram which is used to detect the primary and secondary tempo and their relative strength.In the three part of the article,I introduced the method of the compressed domain tempo detector.firstly,introduce the concept of beat-pattern based error concealment,the concept is quite simple and straightforward.If the lost or distorted segment of the audio signal includes a beat,it would be better to replace it with a segment from a previous beat.In order to reduce the effect of boundary,we introduced a time domain aliasing cancellation technical MDCT(modified discrete cosine transform) .The MDCT transform and inverse transform of signal x(n) are:2)Beat candidate selection.The basic principle of beat candidate selection is setting a proper threshold for the extracted FV. The local maxima within a search window, which fulfils certain conditions, are selected to be beat candidates. This process is performed in each band separately. There are two threshold-based approaches for selecting beat candidates. The first approach uses the primitive FV (multi-band energy) directly and the second approach uses an improved FV(EMR).3)Statistical model.We define a valid candidate in each band as an onset and store a number of previous 101 values in a FIFO buffer for beat prediction in each band. Then we use the median of the 101 vector to calculate the confidence scores of all beat candidates in individual bands.4)Store beat information.Beat position, IBI, and overall confidence score arc sent to the application module after checking with the WSP.The aim of this thesis is to introduce several methods for music tempo detection.The tempo detection of music is an establishment in the multi-disciplinary foundation comprehensive technology.Along with many new theories ,the new method,the new technology unceasing appearances,the content in this field will grow richer and get the development.