The Methods Of Musical Note Recognition

With the development of speech recognition and computer technology, they hope to apply computer technology to recognize the real-time musical sound and then finish memorizing automatically. The technology to let computer understand the sound being played on the musical instrument has important values in composing at the computer. The main tasks of music recognition is to obtain the content of music from musicial audio signal, i. e the score of the music. This system has a broad application prospects in the field Of computer music, Computer-aided composition and digital music for achieving the music recording work with computer expediently. How to realize the recognition of the musical notes is the main research aspect in this paper. Music recognition research has related to music, psychoacoustic, computer, maths, signal processing, Analysis of rhythm and instrument, and pitch extraction and so on.so, this paper gives several algorithms from the perspective of time-frequency analysis.Firstly, we begin with the basic task of music recognition, and propose an algrithm based on HMM. And we focus on the pitch entraction. The key of the recogntion using HMM is to select proper recognition unit. Considering the stability of the feature. The process of recognition is to estimate all statesusing the stream of the features and then we get the best series of states and the series of notes. With the HMM, the recognition of music instruments, the grouping of music classes and so on, all received better results.Secondly, we introduce an algorithm of multipe fundamental frequency estimation, based on the spectral smoothness and the harmonicity. A new method for estimating the fundamental frequencies of concurrent musical sounds is described. The method is based on an iterative approach, where the fundamental frequency of the most prominent sound is estimated, the sound is subtracted from the mixture, and the process is repeated for the residual signal. With these techniques, multiple fundamental frequency estimation can be performed quite accurately in a single time frame, without the use of long-term temporal features.Thirdly, we introduce a new polyphonic pitch estimation method based on SVM. This method is based on machine learning. The basic idea is to consider the polyphonic pitch estimation as a pattern recognition problem. Firstly, the music signal is processed to produce the average energy spectrum by a multi-resolution fast RTFI analysis on the logarithm scale. Finally,the peaks are picked from the average energy spe ctrum to produce the input vector to the SVM. In the training phase, the extracted peaks are used with the target output to produce the SVM polyphonic estimator. Each two-class SVMclassifier recognizes whether an input sample includes the corresponding note or not;and can also output the probability that every music note can occur in the input sample.At last, we introduce a method that using frequency and time-domain information to automatic piano transcript ion.First, we propose a method that groups spectral information in the frequency-domain and uses a rule-based framework to deal with the known problems of polyphony and harmonicit y. Then, we present a novel method for mul tipitch -estimation that uses both frequency and time-domain information. It assumes signal segments to be the linearly weighted sum of waveforms in a database of individual piano notes.The paper gives several algorithms of music recognition.They all complete the note recognition. At last,we compare these methods and show that they all have good aspects and shortcomings at the same time.