CN101650941A - Objective evaluation method for singing tone clearness based on audio frequency spectrum characteristic analysis - Google Patents

Abstract

The invention provides an objective evaluation method for singing tone clearness based on audio frequency spectrum characteristic analysis. The method comprises the following steps: 1) selecting audiosignals of a singing part for tone clearness evaluation; 2) judging singing/interval on the selected audio signals of the singing part; 3) calculating and obtaining measurements of three spectrum characteristics which are energy summation at 2K-3KHz, energy summation of frequency spectrum, and energy summation of frequency weighted spectrum for the singing part obtained in step 2); and 4) calculating and obtaining final objective evaluation parameter indexes which are average energy ratio at 2K-3KHz, average spectrum gravity center according to the frequency spectrum characteristics obtainedin the step 3), and evaluating the singing tone clearness according to the obtained parameter indexes. The evaluation method is characterized by providing objective evaluation parameter indexes that can be quantized by computer algorithms according to the frequency spectrum characteristics analysis of the singing part without influence from human factors, and has the advantages of stability, objectivity and convenience.

Description

Performance tone color lightness objective evaluating method based on the audible spectrum signature analysis

Technical field

The present invention relates to a kind of objective method of singing the tone color lightness of measuring, more particularly, the present invention relates to a kind of objective evaluating method of the performance tone color lightness based on the audible spectrum signature analysis.

Background technology

The tone color lightness is meant a kind of subjective index that is used to estimate singer's performance level, is commonly defined as a kind of bright, high, the warm subjective perception to singer's song.The subjective assessment people feels that singer's song is bright more, high, thinks that then singer's performance tone color lightness is high more, otherwise, he thinks that then the lightness of performance tone color is low more.The subjective assessment people here generally refers to the vocal music expert, such as vocal music teacher and senior musician.And the singer can be singer, the student of vocal music department of specialty, also can be the ordinary people.

At present, sing the stage that relies on expert's subjective assessment fully that substantially still is in of estimating, unripe objective evaluation system can measure and estimate certain aspect of performance level or performance level.The evaluation of singing the bright degree of tone color also relies on expert's subjective perception fully, does not also have the method for objective metric or the objective indicator of auxiliary subjective assessment at present.Though some research find that the evaluation of tone color lightness is relevant with spectrum energy distribution, resonance peak position, energy etc., do not have concrete, well-defined objective metric method appearance abroad.For example, it is relevant with the spectrum energy distribution that document 1 proposes the tone color lightness, but do not provide concrete computing method.In addition, the evaluation of singing level is relevant with the accuracy of melody pitch, the extraction of singer's pitch, and we have adopted the method for proposition in the list of references 2, and the fundamental frequency of this method by asking for every frame data frequency spectrum harmonic wave and maximal value correspondence is as the pitch value of every frame.Here, need to prove that the pitch value that this method obtains is as just a basic parameter value calculating tone color, and is irrelevant with the uniqueness of this patent.

In addition, the subjective evaluation method of singing inevitably has following shortcoming: the first, and nonrepeatability.Because the influence of variety of factors, sing fragment for same one, difference appears in more or less capital between twice subjective assessment, and this species diversity should not exist in the ideal case.The second, subjective skewed popularity.Because each appraiser has different separately hobbies,, skewed popularity is separately arranged all so for same performance fragment, each appraiser has different separately evaluations.The 3rd, the expense costliness, availability is poor.For general ordinary people (not being the student of the song school of specialty), obtain vocal music expert's subjective assessment, be very difficult or costly.

List of references 1:E.Ekholm, G.Papagiannis, and F.Chagnon, " Relating ObjectiveMeasurements to Expert Evaluation of Voice Quality in Western Classical Singing:Critical Perceptual Parameters; " Journal of Voice, vol.12, no.2, pp.182-196,1998.

List of references 2:C.Cao, M.Li, J.Liu, and Y.Yan, " Singing Melody Extraction inPolyphonic Music by Harmonic Tracking; " Proc.8th International Conference on MusicInformation Retrieval (ISMIR), pp.373-374,2007.

Summary of the invention

Therefore,, the object of the present invention is to provide a kind of objective evaluating method of the performance tone color lightness based on the audible spectrum signature analysis,, provide the objective evaluating parameter index that can quantize by to singing the spectrum sigtral response of fragment in view of above-mentioned present situation.

For achieving the above object, the objective evaluating method of the performance tone color lightness based on the audible spectrum signature analysis provided by the invention comprises the steps:

1) chooses the performance fragment sound signal that needs the lightness of tolerance tone color;

2) above-mentioned selected performance fragment sound signal being carried out " performance/gap " judges;

3) to above-mentioned steps 2) in the performance fragment that obtains, calculate three kinds of spectrum signatures and estimate: the 2K-3KHz energy and, the spectrum energy of spectrum energy summation and frequency weighting and;

4) according to above-mentioned steps 3) in the spectrum signature that obtains, calculate the objective metric parameter index of final tone color lightness: average 2K-3KHz energy utilizes the parameter index of obtaining that the lightness of singing tone color is evaluated and tested than, averaging spectrum center of gravity then.

Wherein, above-mentioned steps 4) in, when the parameter index of obtaining in utilization is evaluated and tested the lightness of singing tone color, can be separately based on described average 2K-3KHz energy than or the averaging spectrum center of gravity estimate, also can described average 2K-3KHz energy compare and the averaging spectrum center of gravity combines and estimates.

In addition, the determination methods described step 2) comprises: based on the voice activity detection algorithms of energy and based on the voice/nonvoice judgment mechanism of fundamental detection algorithm.

Compare with the method for existing subjective assessment, two kinds of objective metric methods that the present invention mentions have the following advantages: stability, objectivity, convenience.Owing to be based on the method for objectively evaluating of audio signal frequency spectrum feature,, be not subjected to other artificial factor affecting as long as the performance sound signal unanimity of input is then measured the result and remained unchanged.Simultaneously, the convenience degree of computerized algorithm is improved than the mode of vocal music expert subjective assessment, and the performance fragment that only needs to record is imported, and algorithm promptly can provide the objective metric parameter.

Description of drawings

Fig. 1 is the process flow diagram of tone color lightness objective evaluating method of the present invention.

Embodiment

Below in conjunction with the drawings and the specific embodiments the objective evaluating method of the performance tone color lightness based on the audible spectrum signature analysis of the present invention is done and to be described in further detail.

Fig. 1 is the process flow diagram of tone color lightness objective evaluating method of the present invention.

As shown in Figure 1, the performance tone color lightness objective evaluating method based on the audible spectrum signature analysis provided by the invention comprises the steps:

1) choose the performance fragment sound signal that needs the lightness of tolerance tone color:

In this step 1), at first the voice data of fragment is sung in input.

In the objective evaluating method based on the tone color lightness of singing spectrum sigtral response of the present invention, the form of voice data can adopt any sampling rate such as 16K 16Bit Windows PCM form to handle.As an embodiment, we adopt the audio data format of 16K sampling rate, and each sampled point is represented with the 16Bit number.

2) above-mentioned selected performance fragment sound signal being carried out " performance/gap " judges:

In this " performance/gap " determining step, can adopt several different methods, for example based on voice activity detection (VAD:Voice Activity Detection) algorithm of energy etc.Here, be that example describes to adopt voice/nonvoice judgment mechanism based on the fundamental detection algorithm.Specifically, use based on harmonic wave and the fundamental frequency extraction algorithm, this method is by calculating normalization harmonic wave and spectrum to signal fundamental frequency f ₀Sequence is estimated, fundamental frequency estimation value of each unvoiced frame output.Describe basically identical in the concrete computing method of fundamental frequency and the list of references 2, obtain by the fundamental frequency of asking for every frame data frequency spectrum harmonic wave and maximal value correspondence.The frame that will have effective fundamental frequency value is considered as " performance " part, and remainder is considered as " gap " part.

3) to above-mentioned steps 2) in the performance fragment that obtains, calculate three kinds of spectrum signatures and estimate: the 2K-3KHz energy and, the spectrum energy of spectrum energy summation and frequency weighting and:

By step 2) determined after the performance part position, then carry out the calculating of estimating of spectrum signature (2K-3KHz energy and, the spectrum energy of spectrum energy summation, frequency weighting and).

First feature: the 2K-3KHz energy and.2K-3KHz energy and the summation that refers to 2KHz in each " performance " frame～3KHz partial frequency spectrum energy, its computing method are:

$E_{2K-3K}=\underset{2K}{\overset{3K}{\∫}}{S}_i\left(f\right)\mathrm{df}$

S wherein _i(f) be the fast Fourier transform (FFT) energy spectrum of this frame.

Second feature: the spectrum energy summation, i.e. the energy summation of entire spectrum, its computing formula is:

E _T＝∫S _i(f)df

The 3rd feature: the spectrum energy of frequency weighting and, promptly come the spectrum energy value is weighted with frequency values, its computing method are:

E _W＝∫f·S _i(f)df

4) according to above-mentioned steps 3) in the spectrum signature that obtains, calculate the objective metric parameter index of final tone color lightness: average 2K-3KHz energy than and the averaging spectrum center of gravity.

Obtaining " 2K-3KHz energy and ", " spectrum energy summation " afterwards, calculating average 2K-3KHz energy and liken to singing the objective metric index of tone color lightness.

The calculating of average 2K-3KHz energy ratio:

For any performance frame, the computing formula of its 2K-3KHz energy ratio is:

$R_i=\frac{E_{2K-3K}}{E_T}=\frac{\underset{2K}{\overset{3K}{\∫}}{S}_i\left(f\right)\mathrm{df}}{{\∫S}_i\left(f\right)\mathrm{df}}$

Average 2K-3KHz energy than R exactly in this performance sound signal all performance frame 2K-3KHz energy compare R _iArithmetic mean.In the present invention, average 2K-3KHz energy is than promptly can be used as an objective metric singing the tone color lightness, and span is 0～1, average 2K-3KHz energy is bigger than more, illustrates that the lightness of performance fragment tone color is high more, otherwise, average 2K-3KHz energy illustrates that than more little the tone color lightness is low more.

Perhaps, also can obtained " spectrum energy summation ", " spectrum energy of frequency weighting and " afterwards, calculates the objective metric index that the averaging spectrum center of gravity is used as singing the tone color lightness.

The calculating of averaging spectrum center of gravity:

For any performance frame, its spectrum calculation formula of gravity centre is:

$C_i=\frac{E_W}{E_T}=\frac{\∫f\·S_i\left(f\right)\mathrm{df}}{{\∫S}_i\left(f\right)\mathrm{df}}$

Averaging spectrum center of gravity C promptly is that all sing frame spectrum center of gravity C in this performance sound signal _iArithmetic mean.In the present invention, the averaging spectrum center of gravity also can be used as another objective metric of singing the tone color lightness, and span is 0～8000, the averaging spectrum center of gravity is big more, illustrates that the lightness of performance fragment tone color is high more, otherwise, the averaging spectrum center of gravity is more little, illustrates that the tone color lightness is low more.

In addition, of the present invention based on the tone color lightness objective evaluating method of singing the fragment spectrum sigtral response, above-mentioned two metric can also be combined to tone color lightness and evaluate and test.Be about to two kinds of values of estimating to be weighted after average as index.We do not do concrete qualification as for the mode of weighting and concrete numerical value, can be linear weighted functions, can be the index weightings yet.

Claims (4)

1, a kind of objective evaluating method of the performance tone color lightness based on the audible spectrum signature analysis comprises the steps:

1) chooses the performance fragment sound signal that needs the lightness of tolerance tone color;

2) above-mentioned selected performance fragment sound signal being carried out " performance/gap " judges;

3) to above-mentioned steps 2) in the performance fragment that obtains, calculate three kinds of spectrum signatures and estimate: the 2K-3KHz energy and, the spectrum energy of spectrum energy summation and frequency weighting and;

4) according to above-mentioned steps 3) in the spectrum signature that obtains, calculate the objective metric parameter index of final tone color lightness: average 2K-3KHz energy utilizes the parameter index of obtaining that the lightness of singing tone color is evaluated and tested than, averaging spectrum center of gravity then.

2, the objective evaluating method of the performance tone color lightness based on the audible spectrum signature analysis as claimed in claim 1, it is characterized in that, in the described step 4), when the parameter index of obtaining in utilization is evaluated and tested the lightness of singing tone color, separately based on described average 2K-3KHz energy than or the averaging spectrum center of gravity estimate, perhaps will described average 2K-3KHz energy compare and the averaging spectrum center of gravity combines and estimates.

3, the objective evaluating method of the performance tone color lightness based on the audible spectrum signature analysis as claimed in claim 1 or 2, it is characterized in that described step 2) in determination methods comprise: based on the voice activity detection algorithms of energy and based on the voice/nonvoice judgment mechanism of fundamental detection algorithm.

4, the objective evaluating method of the performance tone color lightness based on the audible spectrum signature analysis as claimed in claim 1 or 2 is characterized in that,

The computing method that each spectrum signature is estimated in the described step 3) are respectively:

1) the 2K-3KHz energy and: computing formula is

$E_{2K-3K}=\underset{2K}{\overset{3K}{\∫}}{S}_i\left(f\right)\mathrm{df}$

Wherein, S _i(f) be the fast fourier transform energy spectrum of this frame;

2) spectrum energy summation: computing formula is

E _T＝∫S _i(f)df；

3) spectrum energy of frequency weighting and:

With frequency values the spectrum energy value is weighted, its computing method are,

E _W＝∫f·S _i(f)df；

The objective metric parameter index of the final tone color lightness that calculates according to above-mentioned spectrum signature in the described step 4) is:

1) average 2K-3KHz energy is than R:

For any performance frame, the computing formula of its 2K-3KHz energy ratio is,

$R_i=\frac{E_{2K-3K}}{E_T}=\frac{\underset{2K}{\overset{3K}{\text{\∫}}}{S}_i\left(f\right)\mathrm{df}}{\∫S_i\left(f\right)\mathrm{df}}$

Described average 2K-3KHz energy is that all performance frame 2K-3KHz energy compare R in this performance sound signal than R _iArithmetic mean, when singing objective metric of tone color lightness, span is 0～1 to this average 2K-3KHz energy than R, average 2K-3KHz energy is bigger more than R, it is high more then to sing the lightness of fragment tone color;

2) averaging spectrum center of gravity C:

For any performance frame, its spectrum calculation formula of gravity centre be,

$C_i=\frac{E_W}{E_T}=\frac{\∫f\·S_i\left(f\right)\mathrm{df}}{\∫S_i\left(f\right)\mathrm{df}}$

Averaging spectrum center of gravity C is that all sing frame spectrum center of gravity C in this performance sound signal _iArithmetic mean, this averaging spectrum center of gravity C is when singing another objective metric of tone color lightness, span is 0～8000, averaging spectrum center of gravity C is big more, it is high more then to sing the lightness of fragment tone color;

3) when above-mentioned average 2K-3KHz energy is combined to tone color than R and two metric of averaging spectrum center of gravity C that lightness is evaluated and tested, as index, the mode of described weighting adopts linear weighted function or exponential weighting with these the two kinds values of estimating to be weighted after average.

Images