Extraction Of The Fundamental Frequency Of Portamento In Polyphonic Music

Portamento means, in the instrumental or vocal performance, the sliding change from one pitch to another pitch. In this process, there is no fracture or gaps in the sound made by performer. Portamento can reflect not only the characteristics and style of music, but also the individuality of the singer or the player. As a result, the parameter estimations of the portamento have potential significance to music classification, singing style, and singer or player recognition, song synthesis and music information annotation.The main goal of this thesis is to extract the trajectory of fundamental frequency, which contains two aspects:the estimation of the fundamental frequency of portamento in monophonic music and polyphonic music. As the musical sound is complicated and the portamento fundamental frequency estimation involves the estimations of both the fundamental frequency and the parameters of nonlinear frequency modulation, in this thesis, portamento is modulated as a set of multi-component nonlinear frequency modulated signals whose frequencies are multiple integer related to each other. Then a variety of time-frequency analysis methods are used to estimate the trajectory of fundamental frequency of this model. According to the characteristic of real portamento, calculate Harmonic Product Spectrum (HPS) based on the pectrogram acquired by those time-frequency analysis methods in order to extrude the fundamental frequency component. To test the performance of the proposed methods for real portamento, we select portamentos created by four different instruments. According to the estimated fundamental frequency, we firstly synthesize a portamento, and then calculate the cross correlation coefficient between the spectrums of synthetic portamento and the real one, in order to compare the accuracy of different methods to different kind of portamentos. At the same time, the Mean Opinion Score (MOS) evaluation is also used to evaluate the fundamental pitch of the synthetic portamento on the estimated fundamental frequency. Thus, both the objective and subjective evaluations are included in comparing the different methods. Experiments show that Smoothed Pseudo Wigner-Ville Distribution (SPWVD) combined with HPS performs better than the others in the estimation of the fundamental frequency of portamento in monophonic music.The course of the portamento fundamental frequency estimation in polyphonic music is built based on the course of the fundamental frequency estimation of monophonic music. Due to that two different portamentos rarely appear at same time in real music, and the energy of portamento is always larger than the other music components, this paper combines time-frequency analysis methods with HPS to extract the fundamental frequency of portamento, and then uses the rotation operator in Polynomial Chirplet Transform (PCT) to demodulate the portamento signal. Then each harmonic of portamento is removed from the spectrum of the portamento signal according to iterative spectral subtraction, so the portamento fundamental frequency can be estimated from the residual signal. In this way, we can get both the portamento fundamental frequency and the pitches of non-portamento components. Since SPWVD cannot back to the original signal by inverse transform, the Short Time Fourier Transform (STFT) combined with HPS is selected to estimate the fundamental frequency of polyphonic music. The experiment results show that the designed course in this thesis is effective for both monophonic and polyphonic music.