Orgin:Using Audio Onset Detection Algorithms
This document only records part of the content , mainly aubio related content-based, not the entire contents of the entire document, record people : [email protected]
QMUL algorithm : This algorithm is based on a signal that combines power ( energy of the observed signal ) and the phase ( observed FFT state deviation ) jointly constitute a complex domain. It comprises an adaptive whitening assembly time and frequency smoothing the changes in the signal, so that by "introducing a similar amplitude dynamic range of each band, so that large amplitude peaks more apparent . The complex algorithm to follow each possibility of occurrence of unexpected events in the frequency domain calculating the peak region, and using a peak selection algorithm to mark onset.
Aubio: About onset detection algorithm aubio and QMUL algorithm similar which improves the onset detection automatic correction functions, by calculating the beat period, Phase phase alignment method . The main prediction is done according to the period, phase, beat. This algorithm has two major variables parameters: threshold threshold 0.01-0.99 ( mainly for peak picking ) and onset mode ( for detection function, comprising a high frequency content, complex domain, and spectral energy difference ) .
For example, the following flute music analysis using the analysis method in the complex domain. Then the system tuning parameter variables FFT bin size is 1024 , the increment size is 512 , the threshold peak threshold is set to 0.5, the silent threshold is set to -50dB, and the minimum inner onset interval is set to 40ms . Since the phase of the presence of the sound encoder, the window size of window is set to 1024 , the number of hops hope to 512. By changing the threshold value of peak selection algorithms, higher or lower, it would lead to too much or too little of onset .
The actual situation and aubio the onset comparative analysis results show the algorithm shown above. In the graph you can see that in this music 11 a true onset it has been correctly found.
Pyin algorithm: it is different from the above-described algorithm is that his aim is to detect the pitch , rather than explicit onset Detection , and is a probability-based approach. It extracts a pitch within a given range of the frequency domain. Because of the additional time stamp of a basic design of the algorithm at the fundamental frequency estimation, it is proved that he is a valid onset detection competitors. At the same time this information can be used to infer the note onset.
Example of use, the system uses FFT bin size of 1024 samples, the increment size is 512 , YIN threshold ( pitch correlation values for a set of candidate probability ) all set 11, to suppress the amplitude of the low pitch estimate is set to 0.1 ( the amplitude suppressed below a certain value ), onset sensitivity is set to 0.7 ( corresponding to peak picking ) , the analysis results as shown below:
Please indicate the source: https: //www.cnblogs.com/lihaiping/p/aubio.html