CN101894560B - Reference source-free MP3 audio frequency definition objective evaluation method - Google Patents

Abstract

The invention relates to a reference source-free MP3 audio frequency definition objective evaluation method, which directly and objectively evaluates the definition quality of MP3 audio frequency. The method comprises the following steps of: firstly, acquiring an MP3 stereo audio frequency file and extracting a medium-high frequency index (MHFI), which affects the definition quality of the audio frequency, from the MP3 stereo audio frequency file, wherein the high audio frequency definition means that the audio frequency comprises more medium-high frequency components; secondly, computing possibilistic entropy of the medium-high frequency index of the audio frequency of each frame, wherein the entropy value reflects the rich degree of the audio frequency information, and the possibilistic entropy value of the highly definite audio frequency is higher than that of the indefinite audio frequency, so that the possibilistic entropy value can be used as a reference source-free audio frequency definition evaluation function; and finally, calculating and mapping the entropy value and defining the value of the audio frequency definition between 0 and 5. Experiments show that the method can effectively measure an objective evaluation value of the reference source-free MP3 stereo audio frequency definition, and the value is close to a subject evaluation value and is consistent with the hearing feeling of human ears.

Description

A kind of MP3 audio frequency definition objective evaluation method of reference source-free

Technical field

The present invention relates to the audio quality objective evaluation, particularly a kind of definition objective evaluation method to reference source-free MP3 (MPEG1-layer3) audio frequency.

Background technology

The audio frequency definition objective assessment technique of a kind of reference source-free of the present invention is that direct definition quality to audio system output signal carries out the score value judge in the 0-5 interval.

The audio quality objective evaluation can be divided into based on the evaluation of input-output with based on the evaluation of exporting from evaluation structure.Wherein most of work concentrates on the evaluation based on input-output, and it is a quality of differentiating voice quality with the error size between the input signal of voice system and the output signal, is a kind of error metrics; And be only to carry out quality assessment, reference source-free according to the output signal of voice system based on the evaluation of output; On the other hand, can be divided on the evaluation content audio frequency whole comprehensive evaluation and branch index (as: sharpness, richness, lightness, soften etc.) are estimated.Audio quality method for objectively evaluating based on input-output is fairly perfect at present; Start late and be to estimate to the audio frequency oeverall quality basically aspect the evaluation content based on the evaluation aspect of output, mainly contain the perception linear prediction method, based on the evaluation method of fuzzy multi-class support vector machine, the method for tolerance sound spectrograph Density Distribution characteristic etc.Also do not relate at present the method that output signal branch index is carried out quality assessment, the articulation index that the present invention studies reference source-free MP3 sound signal just carries out objective evaluation.

The audio frequency definition objective evaluation method of reference source-free proposed by the invention; Improved that subjective assessment is time-consuming, effort and the high shortcoming of cost; Also solved simultaneously occupy superiority at present be difficult to provide the unfavorable factor of reference signal sometimes based on the input-output method for objectively evaluating; The objective evaluation that may further be other indexs such as audio frequency richness, lightness provides reference; Simultaneously also can be used for the speech recognition and the systematic searching of MP3 compression domain with each item index as high-rise perceptual parameters.

Summary of the invention

The object of the present invention is to provide a kind of MP3 audio frequency definition objective evaluation method of reference source-free, the criterion of optimal audio definition quality is provided.From the MP3 packed data, directly extract the characteristic parameter-medium-high frequency index (MHFI) of reflection audio frequency definition; And the possibility entropy of calculating MHFI; Through entropy is added up mapping; The score value of audio frequency definition to be measured is limited between the 0-5 branch, realizes objective evaluation reference source-free audio frequency definition index.

The technical scheme that the present invention solves its technical matters employing is: from the MP3 audio compressed data, extract the medium-high frequency index earlier, calculate the possibility entropy of medium-high frequency index, shine upon the score value that obtains audio frequency definition to be measured through statistics.

The technical solution adopted for the present invention to solve the technical problems can also be further perfect.At first from the MP3 audio compressed data, generate and revise discrete cosine transform MDCT (Modified Discrete Cosine Transform) matrix; Extract effective characteristic parameter again: medium-high frequency index; Calculate the possibility entropy of medium-high frequency index; Through entropy is added up mapping, the score value of audio frequency definition to be measured is limited between 0-5 divides, realize objective evaluation to reference source-free audio frequency definition index.This method specifically comprises the steps:

1) pre-service of MP3 compressed audio: comprise that side information reads to the decoding frame head, master data reads, four parts of Hafman decoding and quantification;

2) generate the MDCT matrix: find out the MDCT coefficient in each subband, coefficient in the subband is arranged, form matrix three parts;

3) extraction of compression domain characteristic parameter: medium-high frequency index M HFI (Medium-high frequency index);

4) the possibility entropy (E) of calculating medium-high frequency index:

5) statistics mapping: the medium-high frequency index possibility entropy that will calculate gained is added up, and it is mapped to the interval and output of 0-5.

The effect that the present invention is useful is: directly from the MP3 audio compressed data, extract effective characteristic parameter, than with extracting characteristic behind the packed data decompress(ion) again, both algorithm was simpler, saved computing time again; Remedied that subjective assessment is time-consuming, defective such as effort and cost height; Also solved simultaneously occupy superiority at present be difficult to provide the unfavorable factor of reference signal sometimes based on the input-output method for objectively evaluating; The objective evaluation that may further be other indexs such as audio frequency richness, lightness provides reference; Also can be used for the speech recognition and retrieval of MP3 compression domain simultaneously with each item index as high-rise perceptual parameters.

Description of drawings

Fig. 1 is the process flow diagram of the MP3 audio frequency definition objective evaluation method of a kind of reference source-free of the present invention.

Fig. 2 is a comparison diagram main, the objective evaluation score value.

Embodiment

A preferred embodiment of the present invention combines accompanying drawing 1 explanation as follows: the MP3 audio frequency definition objective evaluation method of this reference source-free was divided into for five steps:

The first step: MP3 compressed domain audio data processing

The compressed domain audio data processing can be divided into: frame head information reads, the reading of side information, and master data reads, Hafman decoding and quantification.

1) frame head information reads

A) structure of frame head information is deposited in definition;

B) read synchronizing information in the frame;

C) make demoder and synchronization of data streams;

D) confirm this frame data reference position, deposit frame head information;

2) side information reads

A) structure of side information is deposited in definition;

B) confirm the side information starting position by the frame head end position;

C) deposit side information;

3) master data reads

A) structure of zoom factor is deposited in definition, deposits the master data size;

B) calculate master data length;

C) memory headroom of application master data length;

D) read master data;

E) read zoom factor;

4) Hafman decoding and inverse quantization

A) define the array is [32] [18] that deposits the Hafman decoding data in the particle;

B) confirm Huffman data reference position in the master data according to side information;

C) the Huffman data are decoded and decoded data is placed among the is [32] [18];

D) data among the is [32] [18] are carried out inverse quantization, still deposit among the is [32] [18].

Second step: generate the MDCT matrix of coefficients

The data of each particle are made up of 32 subbands and each subband contains 18 coefficients, and according to the principle that frequency distributes from low to high, each particle can form one 32 * 18 matrix.This process is following:

1, finds out each sub-band coefficients

A) find out the coefficient S of subband among the is [32] [18] _i, totally 32;

B) definition S _iCoefficient in the subband is S _i[j], 18 of each sub-band coefficients.

2, form the row vector

A) arrange S again by frequency height principle _iMiddle coefficient is still deposited in S _iIn [j];

B) each subband is arranged the S after accomplishing _i[j] regards the row vector in the matrix as.

3, form matrix

A) with S _i[j] row vector is combined to form 32 * 18M [i] [j] according to sub-band serial number;

B) according to mentioned above principle, the MDCT matrix of coefficients of two particles is expressed as M in the frame ₁[i] [j], M ₂[i] [j]

The 3rd step: the extraction of compression domain characteristic parameter

The compression domain of being extracted is characterized as: medium-high frequency index M HFI (Medium-high frequency index) parameter, and concrete calculation procedure is following:

A) quadratic sum of each particle correction discrete cosine transform coefficient of calculating MP3 audio frequency:

$\underset{i=1}{\overset{32}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right];$

What i, j represented in the formula is the sequence number of coefficient in sideband sequence number and the sideband; M [i] [j] is the MDCT coefficient value.

B) calculate the quadratic sum of each particle medium-high frequency section MDCT coefficient:

$\underset{i=2}{\overset{7}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right];$

Wherein: appropriateness adjustment among a small circle can be carried out according to selected medium-high frequency frequency range in the coefficient index interval;

C) the medium-high frequency index M HFI (Medium-high frequency index) that defines each particle is:

$\mathrm{MHFI}=\frac{\underset{i=2}{\overset{7}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right]}{\underset{i=1}{\overset{32}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right]}$

The 4th step: the possibility entropy E that calculates the medium-high frequency index

$E=-\underset{i=1}{\overset{4N}{\mathrm{\Σ}}}{p}_i\ln p_i$

Wherein N is the total frame number of testing audio, p _iBe the medium-high frequency exponential quantity, because every frame comprises two particles and testing audio is the MP3 dual-channel audio data, so corresponding four the medium-high frequency exponential quantities of every frame;

The 5th step: statistics mapping

A) ask the average EM that puts in order section audio medium-high frequency index possibility entropy:

$\mathrm{EM}=-\frac{1}{4N}\underset{i=1}{\overset{4N}{\mathrm{\Σ}}}{p}_i\ln p_i$

N is the total frame number of audio frequency in the formula, p _iBe the value of medium-high frequency index, because every frame comprises two particles and testing audio is the MP3 stereo audio data, the value of corresponding four the medium-high frequency indexes of every frame;

B) the average EM with medium-high frequency index possibility entropy maps to the interval and output of 0-5;

At first the average of medium-high frequency index possibility entropy being carried out appropriateness amplifies; Get SII; Through the Nonlinear Mapping function SII value is mapped to the 0-5 interval again; Promptly get articulation index value AI (Articulation Index), mapping function wherein commonly used also comprises: chi square function, logarithmic function, intercepting function, window function, threshold function table, many-valued quantization function;

AI＝10/π*arctan(SII)

Final output sharpness score value.

Experimental result

The audio data that this experiment is used is the MP3 stereo data, and its SF is 44.1KHz.Audio data is divided into three groups, and every group is four sections content is identical but voice data that the definition quality subjective sensation reduces successively.According to above-mentioned audio frequency definition Objective Quality Assessment method, can obtain corresponding definition objective and estimate score value, to carry out data comparing result form following with its subjective assessment score value:

Contrast of table 1:MP3 stereo audio definition quality master, its master of objective evaluation score value contrast, objective evaluation score value is as shown in Figure 2.

Can clearly be seen that from reference source-free audio frequency definition master, objective evaluation score value comparison diagram; Reference source-free MP3 audio frequency definition objective evaluation method of the present invention; Can calculate the sharpness score value of respective audio to the MP3 packed data effectively; And very approaching with the subjective assessment score value, meet the auditory perception of people's ear.

Claims (8)

1. the MP3 audio frequency definition objective evaluation method of a reference source-free; It is characterized in that: at first through MP3 compressed audio partial decoding of h is obtained revising discrete cosine transform coefficient; Secondly to these data computation frequency domain medium-high frequency indexes (MHFI), that is: the ratio of the energy of every frame compressed domain audio signal medium-high frequency frequency content and every frame frequency territory gross energy; Select for use the possibility entropy function as not having then, at last the possibility entropy of medium-high frequency index is added up mapping and obtain definition objective evaluation score value with reference to MP3 audio frequency definition evaluation function.

2. the MP3 audio frequency definition objective evaluation method of a kind of reference source-free according to claim 1, it is characterized in that: the concrete operations step is following:

A) pre-service of MP3 compressed audio: the decoding frame head, side information reads, and master data reads, Hafman decoding and quantification;

B) generate correction discrete cosine transform MDCT matrix: find out the correction discrete cosine transform coefficient in each subband, coefficient in the subband is arranged, form matrix;

C) extraction of compression domain characteristic parameter: medium-high frequency index M HFI, that is: the ratio of the energy of every frame compressed domain audio signal medium-high frequency frequency content and every frame frequency territory gross energy:

$\mathrm{MHFI}=\frac{\underset{i=2}{\overset{7}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right]}{\underset{i=1}{\overset{32}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right]}$

I, j represent the sequence number of coefficient in sideband sequence number and the sideband in the formula, and the value of molecule Intermediate Side Band sequence number i can be carried out appropriateness adjustment among a small circle according to selected medium-high frequency frequency range; M [i] [j] is the MDCT coefficient value;

D) the possibility entropy E of calculating medium-high frequency index:

$E=-\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{p}_i\ln p_i$

K represents the number of different components, p in the formula _iThe possibility of representing the i component to take place, be with the difference of Shannon entropy: the possibility entropy does not need to satisfy each component probability of happening sum again and is necessary for 1 constraint;

E) statistics mapping: the possibility entropy that will calculate gained medium-high frequency index is asked all statistics, and it is mapped to the 0-5 interval;

I. ask the average EM of whole section audio medium-high frequency index possibility entropy:

$\mathrm{EM}=-\frac{1}{4N}\underset{i=1}{\overset{4N}{\mathrm{\Σ}}}{p}_i\ln p_i;$

In the formula: N is the total frame number of audio frequency, p _iBe the value of medium-high frequency index, because every frame comprises two particles and testing audio is the MP3 stereo audio data, corresponding four the medium-high frequency indexes of every frame;

Ii. the average EM with medium-high frequency index possibility entropy maps to the 0-5 interval;

At first the average of medium-high frequency index possibility entropy being carried out appropriateness amplifies; Get SII; Through the Nonlinear Mapping function SII value is mapped to the 0-5 interval again; Promptly get articulation index value AI, mapping function wherein commonly used also comprises: chi square function, logarithmic function, intercepting function, window function, threshold function table, many-valued quantization function;

AI＝10/π*arctan(SII)

F) output sharpness score value: the i.e. articulation index AI value that statistics mapping obtains.

3. the MP3 audio frequency definition objective evaluation method of a kind of reference source-free according to claim 2, it is characterized in that: the pretreated concrete implementation method of said step a) MP3 compressed audio is:

A) frame head information reads,

B) reading of side information,

C) master data reads,

D) Hafman decoding and inverse quantization.

4. the MP3 audio frequency definition objective evaluation method of a kind of reference source-free according to claim 2 is characterized in that: said step b) generates the concrete implementation method of MDCT matrix and is:

1), finds out each sub-band coefficients;

A) find out the coefficient of subband in each particle Hafman decoding data array, totally 32;

B) coefficient in i subband of definition is S _i[j], 18 of each sub-band coefficients;

2), form the row vector:

A) arrange i the coefficient in the subband again by frequency height principle, still deposit in S _iIn [j];

B) each subband is arranged the S after accomplishing _i[j] regards the row vector in the matrix as;

3), form matrix

A) with S _i[j] row vector is combined to form 32 * 18 M [i] [j] according to sub-band serial number;

B) according to mentioned above principle, the MDCT matrix of coefficients of two particles is expressed as M in the frame ₁[i] [j], M ₂[i] [j].

5. the MP3 audio frequency definition objective evaluation method of a kind of reference source-free according to claim 2, it is characterized in that: the concrete implementation method of said step c) compression domain characteristic parameter extraction is:

1), medium-high frequency index M HFI (Medium-high frequency index) parameter

A) quadratic sum of each particle correction discrete cosine transform coefficient of calculating MP3 audio frequency:

$\underset{i=1}{\overset{32}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right];$

What i, j represented in the formula is the sequence number of coefficient in sideband sequence number and the sideband; M [i] [j] is the MDCT coefficient value;

B) calculate the quadratic sum of each particle medium-high frequency frequency range MDCT coefficient:

$\underset{i=2}{\overset{7}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right];$

Wherein: appropriateness adjustment among a small circle can be carried out according to selected medium-high frequency frequency range in the coefficient index interval;

C) the medium-high frequency index M HFI that defines each particle is:

$\mathrm{MHFI}=\frac{\underset{i=2}{\overset{7}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right]}{\underset{i=1}{\overset{32}{\mathrm{\Σ}}}\underset{j=1}{\overset{18}{\mathrm{\Σ}}}{M}^2\left[i\right]\left[j\right]}.$

6. the MP3 audio frequency definition objective evaluation method of a kind of reference source-free according to claim 2 is characterized in that: it is following that said step d) is calculated the concrete grammar of possibility entropy of medium-high frequency index:

$E=-\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{p}_i\ln p_i$

Wherein N is the total frame number of testing audio, p _iBe the medium-high frequency index, because every frame comprises two particles and testing audio is the MP3 stereo audio data, corresponding four the medium-high frequency exponential quantities of every frame.

7. the MP3 audio frequency definition objective evaluation method of a kind of reference source-free according to claim 2 is characterized in that: the concrete implementation method of said step e) statistics mapping is:

A) ask the average EM that puts in order section audio medium-high frequency index possibility entropy:

$\mathrm{EM}=-\frac{1}{4N}\underset{i=1}{\overset{4N}{\mathrm{\Σ}}}{p}_i\ln p_i$

N is the total frame number of audio frequency in the formula, p _iBe the value of medium-high frequency index, because every frame comprises two particles and testing audio is the MP3 stereo audio data, the value of corresponding four the medium-high frequency indexes of every frame;

B) the average EM with medium-high frequency index possibility entropy maps to the 0-5 interval;

At first the average of medium-high frequency index possibility entropy being carried out appropriateness amplifies; Get the SII value; Through the Nonlinear Mapping function SII value is mapped to the 0-5 interval again; Promptly get articulation index value AI, mapping function wherein commonly used also comprises: chi square function, logarithmic function, intercepting function, window function, threshold function table, many-valued quantization function;

AI＝10/π*arctan(SII)。

8. the MP3 audio frequency definition objective evaluation method of a kind of reference source-free according to claim 2 is characterized in that: the articulation index AI value that the sharpness score value in the said step f) obtains for the statistics mapping.

Images