A voice training method suitable for bel canto teaching

Abstract

A voice training method suitable for Bel Canto teaching, relates to a voice training method, the principle of which is: 1) collect standard and non-standard Bel Canto high-pitched voice samples respectively, and perform signal processing. 2) Perform power spectrum fitting on the signal, and then extract acoustic parameters that can represent the characteristics of high-pitched voice. 3.) Calculate the probability density function of the standard characteristic parameters, use the quantile of the probability distribution as the scoring standard of the signal, and establish a scoring formula. 4) Calculate the quantile corresponding to the characteristic parameter of the error signal in the probability distribution, and substitute it into the scoring formula to calculate the score. By comparing the voice signals of Bel Canto teachers and Bel Canto beginners, the method uses the fitted power spectrum to extract characteristic parameters, effectively distinguishes scientific voice from incorrect voice, and establishes a scoring evaluation system for voice training, which plays an auxiliary teaching role in Bel Canto voice training.

Description

Technical Field

[0001] The invention relates to a voice training method, in particular to a voice training method suitable for bel canto teaching. Background Art

[0002] As a form of artistic expression, bel canto is loved by music lovers, but the vocal skills of bel canto are relatively strict, and its teaching form is usually face-to-face teaching. Therefore, the improvement of students' vocal skills depends largely on the subjective teaching level of the teacher. And without the guidance of a teacher, it is difficult for beginners to improve their singing skills through their own vocal exercises. There are a large number of bel canto amateurs in China, but except for music colleges, there are few professional bel canto teachers who can teach non-professionals. Summary of the invention

[0003] The purpose of the present invention is to provide a voice training method suitable for bel canto teaching. The method compares the voice signals of bel canto teachers and bel canto beginners, extracts characteristic parameters using the fitted power spectrum, effectively distinguishes scientific voice from incorrect voice, and establishes a scoring and evaluation system for voice training.

[0004] The objective of the present invention is achieved through the following technical solutions:

[0005] A voice training method suitable for bel canto teaching, the method comprising the following steps:

[0006] 1) Collect high-pitched signals from a certain bel canto beginner and teacher, and remove the noise and silence segments in the sound samples to obtain interference-free bel canto signals;

[0007] 2) Fitting the power spectrum of the processed beautiful voice signal to obtain power spectrum images of the two beautiful voice signals; dividing the power spectrum image into several energy regions, and extracting characteristic parameters of the energy regions as acoustic features of the beautiful voice signal;

[0008] 3) Taking the characteristic parameters of scientific vocalization signals as the criteria for judging high-pitched vocalization, the probability density function and scoring formula of the standard parameters are established;

[0009] 4) Calculate the quantile corresponding to the error signal characteristic parameter in the probability distribution, and substitute it into the scoring formula to calculate the score.

[0010] The method for vocal training suitable for bel canto teaching, in step 2), in order to avoid the phase loss caused by the traditional Fourier transform to the signal, adopts a modern spectrum estimation method, and uses the Burg method in the autoregressive (AR) model method to fit the power spectrum. The Burg method does not need to estimate the autocorrelation function, but uses the known sequence x(n) to calculate the reflection coefficient K k , the reflection coefficient calculation formula is as follows:

[0011]

[0012] Among them, x(n)=e0(n)=b0(n), 0≤n≤N-1 is the initial condition, akk(k=1,2,3...p) is the AR parameter, and p is the AR model order;

[0013] Then use the Levinson recursive algorithm and substitute the K of the p-order AR model k and K of the p-1 order AR model k , calculate a ki , forward prediction error e k (n) and the backward prediction error b k (n), the calculation formula is as follows:

[0014] a ki =a k-1,i +a kk a k-1,k-i

[0015] e k (n) = e p-1 (n)+K p b p-1 (n-1)

[0016] b k (n) = b p-1 (n)+K p e p-1 (n)

[0017] Finally, the above steps are repeated until the expected order is reached, and the AR model parameters of all orders are obtained.

[0018] The voice training method suitable for bel canto teaching, in the step 2), the power spectrum image is divided into several intervals as a whole, as the energy zone of the power spectrum of the voice signal, specifically: 0-6kHz is the first energy zone, 6-11kHz is the second energy zone, 11-15kHz is the third energy zone; the second and third energy zone images of the power spectrum are respectively subjected to first-order fitting and second-order fitting based on the least squares method to obtain a quadratic curve equation and a linear straight line equation; wherein the quadratic term coefficient a of the curve and the slope k of the straight line are key acoustic feature parameters for distinguishing soprano voice signals.

[0019] The voice training method suitable for bel canto teaching, in the step 3), in the characteristic parameters of the power spectrum energy area, the parameters of scientific voice are used as the scoring standard values ​​of high notes, the mean μ and standard deviation σ of the standard parameter set are calculated, and a probability density function is established. The density function formula is as follows:

[0020]

[0021] Calculate the characteristic parameter X of the error signal and its corresponding quantile in the probability distribution If the quantile Y>0.5, then let Y=1-Y, and finally substitute the quantiles Y(a) and Y(k) of the two characteristic parameters into the scoring formula to calculate and obtain the score Z of the erroneous sound signal; the scoring formula is as follows:

[0022] Z=[Y(a)+Y(k)]×100.

[0023] The advantages and effects of the present invention are:

[0024] The present invention proposes a scoring method suitable for high-pitched voice training based on the extraction of professional characteristics of bel canto voice production, which is used to assist bel canto teaching. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of the present invention;

[0026] Figure 2 This is a time domain diagram of an original soprano signal according to an embodiment of the present invention;

[0027] Figure 3 is a time domain diagram of a soprano signal after preprocessing according to an embodiment of the present invention;

[0028] Figure 4 This is a power spectrum diagram of a soprano signal according to an embodiment of the present invention;

[0029] Figure 5 It is a diagram showing the division and comparison of energy zones of a power spectrum image according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The present invention will be described in detail below with reference to the embodiments shown in the accompanying drawings.

[0031] The present invention uses the soprano voice of a bel canto teacher as a criterion, calculates the statistical value of a standard data set, establishes a probability density function, and then performs quantile calculation and scoring on the characteristic parameters of an erroneous bel canto signal. The method comprises: first, collecting soprano signals of bel canto beginners and teachers of a certain conservatory of music, and removing noise segments and silent segments in the sound samples to obtain an interference-free bel canto signal. Secondly, fitting the power spectrum of the processed bel canto signal to obtain a power spectrum image, dividing the power spectrum image into a number of energy zones, and extracting the characteristic parameters of the energy zones as the acoustic characteristics of the bel canto signal. Thirdly, taking the characteristic parameters of the scientific voice signal as a criterion, a probability density function and a scoring formula are established. Finally, the quantile corresponding to the characteristic parameters of the erroneous signal in the probability distribution is calculated, and the score is calculated by substituting them into the scoring formula.

[0032] Example

[0033] like Figure 1-Figure 5 As shown, the specific operations of the present invention are as follows:

[0034] The first step is to collect, screen and classify soprano signals from bel canto beginners and bel canto teachers who have no recent cold symptoms and good voices in a music college, and obtain a series of original soprano vocal signals, such as Figure 2 As shown; Next, the obtained sound signal is preprocessed to remove the noise and silence segments in the sound samples one by one to obtain an interference-free beautiful sound signal, such as Figure 3 shown.

[0035] The second step is, Figure 4 As shown, the processed melody signal is fitted with a power spectrum. In order to avoid the defects brought by the traditional Fourier transform, the present invention adopts the Burg method in the modern spectrum estimation method to fit the power spectrum of the signal. The Burg method does not need to estimate the autocorrelation function, but uses the known sequence x(n) to calculate the reflection coefficient K k , the reflection coefficient calculation formula is as follows:

[0036]

[0037] Among them, x(n)=e0(n)=b0(n), 0≤n≤N-1 is the initial condition, akk(k=1,2,3...p) is the AR parameter, and p is the AR model order.

[0038] Then use the Levinson recursive algorithm and substitute the K of the p-order AR model k and K of the p-1 order AR model k , calculate a ki , forward prediction error e k (n) and the backward prediction error b k (n), the calculation formula is as follows:

[0039] a ki =a k-1,i +a kk a k-1,k-i

[0040] e k (n) = e p-1 (n)+K p b p-1 (n-1)

[0041] b k (n) = b p-1 (n)+K p e p-1 (n)

[0042] Finally, repeat the above steps until the expected order is reached, and the AR model parameters of all orders are obtained. Using the Burg method for power spectrum fitting can reduce the error and obtain the real spectrum estimation effect, and finally obtain the power spectrum image of the soprano voice signal. The soprano signal of the Bel Canto teacher is used as the standard scientific voice and compared with the voice signal of the Bel Canto beginner.

[0043] The third step is Figure 5 As shown in the figure, the power spectrum image is divided into three energy zones, namely 0-6kHz is the first energy zone, 6-11kHz is the second energy zone, and 11-15kHz is the third energy zone. The first-order fitting and second-order fitting based on the least squares method are performed on the second and third energy zone images of the power spectrum, respectively, to obtain a quadratic curve equation and a linear straight line equation. Among them, the quadratic term coefficient a of the curve and the slope k of the straight line are the key acoustic feature parameters used to distinguish the right and wrong soprano vocal signals.

[0044] The fourth step is to use the parameters of scientific voice production as the standard values ​​for soprano scoring, calculate the mean μ and standard deviation σ of the standard parameter set, and establish a probability density function. The density function formula is as follows:

[0045]

[0046] Calculate the characteristic parameter X of the error signal and its corresponding quantile in the probability distribution If the quantile Y>0.5, then let Y=1-Y, and finally substitute the quantiles Y(a) and Y(k) of the two characteristic parameters into the scoring formula to calculate the score Z of the wrong sound signal. The scoring formula is as follows:

[0047] Z=[Y(a)+Y(k)]×100.

Claims

1. A voice training method suitable for bel canto teaching, characterized in that: The method comprises the following steps: 1) Collect high-pitched signals from a certain bel canto beginner and teacher, and remove the noise and silence segments in the sound samples to obtain interference-free bel canto signals; 2) Fitting the power spectrum of the processed beautiful voice signal to obtain power spectrum images of the two beautiful voice signals; dividing the power spectrum image into several energy regions, and extracting characteristic parameters of the energy regions as acoustic features of the beautiful voice signal; 3) Taking the characteristic parameters of scientific vocalization signals as the criteria for judging high-pitched vocalization, the probability density function and scoring formula of the standard parameters are established; 4) Calculate the quantile corresponding to the characteristic parameter of the error signal in the probability distribution and substitute it into the scoring formula to calculate the score; In the step 2), in order to avoid the phase loss of the signal caused by the traditional Fourier transform, a modern spectrum estimation method is used to fit the power spectrum using the Burg method in the AR model method. The AR model method is an autoregressive model method. The Burg method does not need to estimate the autocorrelation function, but uses the known sequence x(n) to calculate the reflection coefficient K k , the reflection coefficient calculation formula is as follows: Among them, x(n)=e0(n)=b0(n), 0≤n≤N-1 is the initial condition, a kk is the AR parameter, k=1, 2, 3...p; p is the AR model order; Then use the Levinson recursive algorithm and substitute the K of the p-order AR model k and K of the p-1 order AR model k , calculate a ki , forward prediction error e k (n) and the backward prediction error b k (n), the calculation formula is as follows: a ki =a k-1，i +a kk a k-1，k-i e k (n)=e p-1 (n)+K p b p-1 (n-1) b k (n)=b p-1 (n)+K p e p-1 (n) Finally, step 2) is repeated until the expected order is reached, and the AR model parameters of all orders are obtained; In the step 2), the power spectrum image is divided into several intervals as a whole, which are energy zones of the power spectrum of the vocal signal, specifically: 0-6kHz is the first energy zone, 6-11kHz is the second energy zone, and 11-15kHz is the third energy zone; the second and third energy zone images of the power spectrum are respectively subjected to first-order fitting and second-order fitting based on the least squares method to obtain a quadratic curve equation and a linear straight line equation; wherein the quadratic term coefficient a of the curve and the slope k of the straight line are key acoustic characteristic parameters for distinguishing soprano vocal signals; In step 3), in the characteristic parameters of the power spectrum energy area, the parameters of scientific voice are used as the scoring standard values ​​of high notes, the mean μ and standard deviation σ of the standard parameter set are calculated, and a probability density function is established. The density function formula is as follows: Calculate the characteristic parameter X of the error signal and its corresponding quantile in the probability distribution If the quantile Y>0.5, then let Y=1-Y, and finally substitute the quantiles Y(a) and Y(k) of the two characteristic parameters into the scoring formula to calculate and obtain the score Z of the erroneous sound signal; the scoring formula is as follows: Z=[Y(a)+Y(k)]×100.

Images