KR20220129878A - Method for Discriminating AMR Voice Data and Apparatus thereof - Google Patents

Abstract

The present invention relates to a method for discriminating adaptive multi-rate (AMR) voice data to easily determine whether voice data is AMR voice data encoded according to an AMR format, and an apparatus thereof. According to one embodiment of the present invention, the method for discriminating AMR voice data comprises: a target data processing step of extracting a centroid, which is a unique factor of target data, from the target data compressed in an AMR format of one of 0^th to seventh modes; a reference data processing step of extracting a reference Eigenfactor, which is an Eigenfactor of reference data, from the reference data compressed in the AMR format; and a comparison and determination step of comparing the centroid and the reference Eigenfactor by using a maximum likelihood (ML) method to discriminate whether the target data is AMR data compressed in an AMR format. The comparison and determination step performs discrimination by a mean squared error (MSE) distribution which is a probability density function of an MSE between the centroid and the reference Eigenfactor.

Description

Method for Discriminating AMR Voice Data and Apparatus thereof

The present invention relates to a method and apparatus for determining AMR voice data, and more particularly, to a method and apparatus for easily determining whether target data is AMR voice data by comparing target data and reference data.

Adaptive Multi-Rate (hereinafter, "AMR") is a patented audio data compression optimized for voice encoding, and the AMR codec is known as a codec that can flexibly cope with a flexible environment by supporting various data rates. AMR encodes a total of 8 variable data rates of 4.75, 5.15, 5.90, 6.70, 7.40, 7.95, 10.2, and 12.2 kbit/s of narrowband signals (200-3400 Hz). It is a multi-rate narrowband voice codec that variably adjusts the output mode to guarantee the user's quality of service (QoS).

Embodiments of the present invention are to provide a method and apparatus for easily determining whether voice data is AMR voice data encoded according to the AMR format through comparison with previously secured reference data.

A target data processing step of extracting a centroid, which is a unique factor of the target data, from the target data compressed in the AMR format of one of the 0th mode to the 7th mode according to an embodiment of the present invention; a reference data processing step of extracting a reference intrinsic factor, which is an intrinsic factor of the reference data, from the reference data compressed in the AMR format; and a comparison determination step of determining whether the target data is AMR data compressed in an AMR format by comparing the centroid and the reference intrinsic factor using a Maximum Likelihood (ML) method, wherein the comparison The determination step is determined using the MSE distribution, which is a probability density function of a mean squared error (MSE) between the centroid and the reference eigen factor.

The target data processing step may include: extracting a principal component from the target data; and extracting the centroid for a partial region of the main component from the main component, wherein the extracting of the centroid comprises dividing the main component into a plurality of subframes including at least one or more bits; and specifying the centroid based on bit information included in some subframes of the plurality of subframes.

An input mode in which the target data is a compressed AMR mode, and a check mode providing the number of bits for extracting the principal component with respect to the input mode include the 0th mode to the 7th mode, and extracting the principal component The step may include: extracting a principal component of the matching mode extracted based on the same check mode as the input mode with respect to the input mode; and extracting, with respect to the input mode, a cross factor, which is a main component of a mismatch mode extracted based on a check mode different from the input mode.

The extracting of the main component may include: a data receiving step of receiving a bitstream of target data compressed in one AMR format among the 0th mode to the 7th mode; a frame separation step of dividing the bitstream into frames having a predetermined first number of bits; a bit array generating step of generating a bit array by integrating the separated frames; and specifying the main component of the bitstream based on the bit array and the total number of frames, wherein the centroid may be extracted based on some of the frames.

The reference data processing step may include: extracting a reference principal component from the reference data; and extracting the reference eigen factor for a partial region of the reference principal component from the reference principal component, wherein the extracting of the reference eigen factor comprises a plurality of subframes including the reference principal component at least one bit or more. and specifying the reference intrinsic factor based on bit information included in some subframes of the plurality of subframes.

The comparison determination step may include: obtaining an MSE distribution for each mode based on the centroid and the reference intrinsic factor; obtaining a threshold value for each mode based on the MSE distribution for each mode; and comparing the minimum value of the MSE distribution of the target data and the threshold value of the input mode among the threshold values for each mode, and determining that the target data is AMR voice data if the minimum value is less than the threshold value. can

The acquiring of the threshold may include acquiring, as a threshold, a point at which the first MSE distribution of the principal component of the inconsistent mode has the same value as the second MSE distribution of the principal component of the inconsistent mode.

A program for executing the method according to the above-described embodiments of the present invention may be stored in a computer-readable recording medium.

a target data processing unit for extracting a centroid, which is a unique factor of the target data, from the target data compressed in the AMR format of one of the 0th mode to the 7th mode according to an embodiment of the present invention; a reference data processing unit for extracting a reference intrinsic factor, which is an intrinsic factor of the reference data, from the reference data compressed in the AMR format; and a comparison unit that compares the centroid and the reference intrinsic factor using a Maximum Likelihood (ML) method to determine whether the target data is AMR data compressed in an AMR format; The negative is determined using the MSE distribution, which is a probability density function of the mean squared error (MSE) between the centroid and the reference eigen factor.

The target data processing unit may include: a principal component extractor configured to extract a principal component from the target data; and a centroid extraction unit for extracting the centroid for a partial region of the main component from the main component, wherein the centroid extraction unit divides the main component into a plurality of subframes including at least one or more bits, the plurality of The centroid may be specified based on bit information included in some subframes of subframes of .

An input mode in which the target data is a compressed AMR mode, and a check mode for providing the number of bits for extracting the principal component with respect to the input mode include the 0th mode to the 7th mode, wherein the principal component extraction unit comprises: For the input mode, a principal component of the coincidence mode extracted based on the same check mode as the input mode is extracted, and with respect to the input mode, a cross factor which is a main component of the inconsistency mode extracted based on a check mode different from the input mode ) can be extracted.

The principal component extracting unit may include: a data receiving unit configured to receive a bitstream of target data compressed in an AMR format of one of the 0th mode to the 7th mode; a frame dividing unit dividing the bitstream into frames having a predetermined first number of bits; a bit array generator for generating a bit array by integrating the separated frames; and a principal component specifying unit for specifying the main component of the bitstream based on the bit array and the total number of frames, wherein the centroid may be extracted based on some of the frames.

The reference data processing unit may include: a reference principal component extractor configured to extract a reference principal component from the reference data; and a reference eigenfactor extracting unit for extracting the reference eigenfactor for a partial region of the reference principal component from the reference principal component, wherein the reference eigenfactor extractor includes a plurality of bits including at least one bit for the reference principal component. It may be divided into subframes, and the reference intrinsic factor may be specified based on bit information included in some subframes of the plurality of subframes.

The comparison unit may include: an MSE acquisition unit for each mode acquiring an MSE distribution for each mode based on the centroid and the reference intrinsic factor; a threshold value acquisition unit for acquiring a threshold value for each mode based on the MSE distribution for each mode; and a comparison determining unit that compares the minimum value of the MSE distribution of the target data and the threshold value of the input mode among the threshold values for each mode, and determines that the target data is AMR voice data when the minimum value is less than the threshold value can do.

The threshold value obtaining unit may obtain, as a threshold, a point at which the first MSE distribution of the principal component of the coincident mode has the same value as the second MSE distribution of the principal component of the inconsistent mode.

According to embodiments of the present invention, it is possible to easily determine whether the voice data is AMR voice data encoded according to the AMR format by comparing it with pre-secured reference data.

1 is a block diagram schematically showing the configuration of an AMR voice data discrimination apparatus according to an embodiment of the present invention.
FIG. 2 is a block diagram schematically illustrating a part of the configuration of the apparatus according to FIG. 1 .
3 is a view for explaining a data processing process by the principal component extractor according to an embodiment of the present invention.
4 is a diagram schematically illustrating one frame and subframes included in the frame in one mode according to an embodiment of the present invention.
5 to 12 are graphs illustrating an example of a graph of main components of male and female voice data encoded in mode 0 to mode 7 of the present invention.
13 and 14 are diagrams for explaining a cross factor according to an embodiment.
15 is a diagram illustrating a matrix for each mode generated by matching an input mode in which target data is encoded and a check mode comparing the same according to an embodiment of the present invention.
16 is a graph illustrating an MSE distribution according to an embodiment of the present invention with respect to mode 0 to mode 3;
17 is a graph illustrating an MSE distribution according to an embodiment of the present invention with respect to a fourth mode to a seventh mode.
18 is a flowchart illustrating a method for determining AMR voice data according to an embodiment of the present invention.

Since the present invention can apply various transformations and can have various embodiments, specific embodiments are illustrated in the drawings and described in detail in the detailed description. Effects and features of the present invention, and a method for achieving them, will become apparent with reference to the embodiments described below in detail in conjunction with the drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various forms.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, and when described with reference to the drawings, the same or corresponding components are given the same reference numerals, and the overlapping description thereof will be omitted. .

In the following embodiments, terms such as first, second, etc. are used for the purpose of distinguishing one component from another, not in a limiting sense. In the following examples, the singular expression includes the plural expression unless the context clearly dictates otherwise. In the following embodiments, terms such as include or have means that the features or components described in the specification are present, and the possibility that one or more other features or components may be added is not excluded in advance. In the drawings, the size of the components may be exaggerated or reduced for convenience of description. For example, since the size and shape of each component shown in the drawings are arbitrarily indicated for convenience of description, the present invention is not necessarily limited to the illustrated bar.

In the following embodiments, when a film, region, or component is connected, other films, regions, and components are interposed between the films, regions, and components as well as when the films, regions, and components are directly connected. It also includes cases where it is indirectly connected. For example, in the present specification, when it is said that a film, a region, a component, etc. are electrically connected, not only the case where the film, a region, a component, etc. are directly electrically connected, but also other films, regions, and components are interposed therebetween. Indirect electrical connection is also included.

1 is a block diagram schematically illustrating the configuration of an AMR voice data discrimination device 10 (hereinafter, may be referred to as a 'discrimination device') according to an embodiment of the present invention.

The determination apparatus 10 may include a target data processing unit 100 , a reference data processing unit 200 , and a comparison unit 300 . Each component constituting the device of FIG. 1 may be a physical unit manufactured to perform a specific function, or may be a logical module implemented as a program and set to implement each function.

The target data processing unit 100 may include a principal component extraction unit 110 and a centroid extraction unit 120 for extracting a centroid, which is a unique factor of the target data, from the principal component. Hereinafter, 'target data' refers to data that is to be estimated whether it is AMR data or not in the present invention. The main component extraction unit 110 will be described in more detail with reference to FIG. 2 to be described later, and the centroid extraction unit 120 will be described in more detail with reference to FIGS. 4 to 13 to be described later.

The reference data processing unit 200 may include a principal component extraction unit 210 and a codeword extraction unit 220 for extracting a codeword that is a unique factor of the reference data from a principal component of the reference data. Hereinafter, 'reference data' refers to sample AMR data to be compared when analyzing the target data.

The comparison unit 300 includes a mode-specific MSE acquisition unit 310, a threshold value acquisition unit 320, and the MSE distribution for obtaining a mean squared error (Mean Squared Error; hereinafter referred to as 'MSE') distribution for each mode. and a comparison determination unit 330 for determining whether the target data is AMR voice data by comparing the threshold value with the minimum value of . In this case, the MSE acquisition unit 310 for each mode may receive the centroid from the target data processing unit 100 and the intrinsic factor from the reference data processing unit 200 to obtain the MSE distribution for each mode. The MSE distribution for each mode will be described in more detail with reference to FIGS. 15 to 17 to be described later.

FIG. 2 is a block diagram schematically illustrating the principal component extraction unit 110 which is a part of the configuration of the apparatus 10 according to FIG. 1 . Hereinafter, the main component extraction unit will be described based on the configuration of the target data processing unit 100, but only the data processed by the principal component extraction unit 210 of the reference data processing unit 200 is different as reference data. can have

The principal component extracting unit 110 may include a data receiving unit 111 , a frame separating unit 112 , a bit array generating unit 113 , and a principal component specifying unit 114 .

The data receiver 111 may receive a bitstream of voice data encoded in one of the 0th mode to the 7th mode in the AMR format. The AMR voice data of the present invention has the number of bits allocated for each mode in which the corresponding voice data is encoded. AMR voice data may include individual compression format features for a total of eight bit rates (variable bit rates), which are variable rate compression options.

More specifically, the raw data input for AMR encoding includes 160 samples per frame, and 50 frames are input every second, but in the present invention, a file for which AMR encoding is already completed is received, Since a method for analyzing a received voice file is proposed, a detailed description thereof will be omitted. The present invention proposes a method for accurately estimating AMR voice data when there is no information on the compression format of the received data even though it is received.

For example, in [Table 1], a mode of 10.2 kbit/s is called mode 6, and when a voice sample is encoded (encoded) in this mode, a bitstream of 204 bits per frame is generated. do. Here, 26 bits among 204 bits are allocated with the exponent of the LSF submatrix, and the remaining 178 bits may be divided into subframes of 46, 43, 46, and 43 bits, respectively. As another example, a mode of 4.75 kbit/s is called mode 0, and a bitstream of 95 bits per frame is generated. For clearer description, modes 0 to 7 will be referred to as modes 0 to 7 hereinafter.

The data receiving unit 111 receives the bitstream of voice data compressed in one of the AMR formats from the 0th mode to the 7th mode, and the present invention assumes that all mode information in the AMR file is the same. For example, when the data receiving unit 111 receives the AMR file, the AMR file contains only the information of the 0th mode or only the information of the 4th mode, and in the analysis result of the file, the first The information of the mode and the second mode is not detected at the same time. The above conditions take into account the fact that codec changes do not occur frequently in actual WCDMA channels.

The frame dividing unit 112 divides the bitstream of the voice data received by the data receiving unit 111 into frames having a predetermined first number of bits. The frame separator 112 may store the information shown in Table 1 and use it when dividing the bitstream into a plurality of frames. The first number of bits is a preset value and means the number of bits included in one frame of each mode in Table 1. The first number of bits may be any one selected from among 95, 103, 118, 134, 148, 159, 204, and 244.

For example, if a bitstream composed of 1020 bits is received and the frame separation unit 112 separates the bitstream into a frame of a fifth mode, the first number of bits becomes 204, and the bitstream is a total of five fifth mode. It can be divided into mod frames. When the final analysis result is calculated according to the first selected number of bits, the frame dividing unit 112 may divide the bitstream into a plurality of frames based on the new first number of bits if an appropriate result is not obtained. have. In the present invention, since the received bitstream does not contain mode information, the repetitive process as described above is analyzed for all modes, and the result of mode analysis showing the most ideal result is analyzed in the present invention. to take

The bit array generator 113 generates a bit array by integrating the separated frames.

Here, the bit array refers to processed data processed through a series of processes because the bitstream of the received voice data is serially elongated data and it is difficult to extract features. The bit array generating unit 113 receives a plurality of frames constituting the voice data from the frame separating unit 112, and based on the first number of bits corresponding to the size of the frame, information included in the plurality of frames Create a bit array containing all of them.

The main component specifying unit 114 specifies the main component based on the total number of the bit array and frames generated by the bit arrangement generating unit 113, and the main component will be described in more detail with reference to FIG. 3 to be described later.

3 is a view for explaining a data processing process by the principal component extractor 110 according to an embodiment of the present invention.

In FIG. 3, the bitstream of voice data consists of a total of (M×N) bits, and is divided into N frames including M bits per frame. The bit array generator 113 may generate a bit array v _s based on N frames. In FIG. 3 , the bit array generating unit 113 generates the bit array by summing the first, second, and third bits of each frame, but according to an embodiment, each bit value of the bit array is not simply summed but through another operation. can be determined.

In addition, since the bit array may be generated by adding several frames, it may be composed of a larger number of bits than the frames generated by the frame dividing unit 112 . For example, in FIG. 3 , the v ₁ frame is a frame including M bits, but the bit array v _s may include (N×M) bits.

When the bit array v _s is generated, the principal component specifying unit 114 divides it into values included in the bit array by N, which is the total number of frames, to specify a principal component of the target voice data. The value included in the bit is 0 or 1, and the principal component includes the result of _dividing the _sum of bits of each frame by N, the total number of frames. When changing from 1 to N, it can be seen that bi is a probability distribution function of 1.

Equation 1 represents the mathematical meaning of the principal component. In Equation 1, N is the total number of frames separated from the bitstream, and v _k means the k-th frame. Also, E[v] in Equation 1 indicates that the principal component specified by the principal component specifying unit 114 can be expressed as a vector. Since the main component is generated by integrating a plurality of frames composed of the first number of bits, the length of the main component may be equal to the first number of bits or equal to n multiples of the first number of bits (in this case, n is a natural number greater than 1).

It will be apparent to those skilled in the art that the number M of bits included in one frame described in FIG. 3 can also be applied to various values (95 or 204, etc.) in [Table 1].

The main component specified as above is a candidate for the centroid extracted by the centroid extraction unit 120 . Here, the fact that the main component becomes a candidate for the intrinsic factor means that the information estimator 150 grasps the compression format information of the voice data through only some of the information included in the main component.

The centroid extraction unit 120 may divide the main component into a plurality of subframes including at least one or more bits. Thereafter, the centroid may be specified based on bit information included in some subframes of the plurality of subframes.

For example, the centroid extractor 120 may divide the main component into first to fourth subframes including at least one bit or more. The target data processing unit 100 stores Table 1 and when the main component is specified by the main component extraction unit 110 , the centroid extraction unit 120 may divide the main component into four subframes. Since the principal component is specified by integrating and averaging the bit values of frames based on Equation 1, it has the same number of bits as the frame having the first number of bits, and has the same number of subframes as the subframes included in the frames. It may include a frame. In other words, the main component may be calculated as an average bit value included in the entire frame as shown in Equation 1 above.

4 is a diagram schematically illustrating one frame and subframes included in the frame in the sixth mode.

Referring to FIG. 4 , it can be seen that in the sixth mode, one frame includes 204 bits, the first 26 bits are the LSP, and the other 178 bits are composed of the first subframes to the fourth subframes. Each bit constituting the frame is labeled from s1 to s204. Specifically with reference to Table 1, 8 bits (s27 to s34) of the first subframe are allocated an exponent of the adaptive codebook, and algebraic codebook information is allocated to 31 bits (s35 to s72). A codebook gain is allocated to the last 7 bits (s66 to s72) of the first subframe. Subsequently, the index of the adaptive codebook is allocated to 5 bits (s73 to s77) of the second subframe, and the algebraic codebook information is allocated to the remaining 38 bits (s78 to s115). Information is allocated to the third subframe and the fourth subframe in the same manner as the first subframe and the second subframe, respectively.

The MSE acquisition unit 310 for each mode of the comparison unit 300 may determine whether the target data is AMR voice data based on bit information included in the third subframe and the fourth subframe. This is because the bit information included in the third subframe and the fourth subframe can function as an eigenfactor in that it is information indicating compressed format information of voice data. In this case, the intrinsic factor of the target data may be defined as a centroid, and the intrinsic factor of the reference data may be defined as a codeword.

mode 3rd subframe 4th subframe 0 s62 to s82 s83 to s95 One s66 to s84 s85 to s103 2 s73 to s97 s98 to s118 3 s81 to s109 s110 to s134 4 s88 to s119 s120 to s148 5 s94 to s127 s128 to s159 6 s116 to s161 s162 to s204 7 s142 to s194 s195 to s244

[Table 2] is a table showing the bit positions of eigen factors in the main component. Specifically, [Table 2] shows the positions of the third subframe and the fourth subframe in the main component composed of the same first number of bits as one frame. Whether the target data is AMR voice data can be determined only with bit information about the third subframe and the fourth subframe.

5 to 12 are graphs illustrating an example of a graph of main components of male and female voice data encoded in mode 0 to mode 7 of the present invention.

More specifically, FIGS. 5 to 12 show that raw voice data recorded while a man and a woman read a document of the same content for 1 minute are encoded into modes 0 to 7, and through the above-described process The principal components are specified and presented in a graph. The blue graph means the average beat value (principal component) for male voice, and the red graph means the average beat value for female voice.

First, FIG. 5 shows a graph of principal components in the 0th mode, the coordinates of the horizontal axis range from 0 to 95 depending on the number of bits in the 0th mode, and values 0 to 1 are assigned to the vertical axis.

Referring to FIG. 5 , it can be seen that the graph for male voice and the graph for female voice in the third and fourth subframes show extremely similar shapes, which are This is a remarkable result when comparing the graph for male voice and the graph for female voice in the subframe and the second subframe section showing a marked difference. That is, the voice difference due to the gender difference is reflected in both the first and second subframe sections, and since both men and women read documents printed with the same characters, the third and fourth subframe sections It may be understood that the correspondence of the graphs in ? includes information about the 0th mode.

Next, FIG. 6 shows a graph of principal components in the first mode. The coordinates of the horizontal axis of FIG. 6 range from 0 to 103 according to the number of bits in the first mode, and values from 0 to 1 are assigned to the vertical axis. In FIG. 6 , it can be seen that the two graphs show similar tendencies in the third subframe and the fourth subframe period, irrespective of gender, as in FIG. 5 .

In addition, in the graphs shown in FIGS. 7 to 12, in the third subframe and the fourth subframe section, the graph for the male voice and the graph for the female voice show extremely similar tendencies, so the comparator 300 ) (in particular, the comparison determination unit 330) analyzes the AMR file whose encoding information is unknown based on the information on this part, so that it is possible to accurately estimate the compression format information.

If the main component specified through the above-described process is not suitable for analysis, the comparator 300 may change the first number of bits to control a new main component to be generated. The frame separating unit 112 divides the bitstream of the voice data into frames of a mode that does not match the original mode, and when a main component is specified after a bit array is generated from the separated frames, the graph of the main component is difficult to analyze. It may be output in an inappropriate form, and such a main component may be referred to as a cross-factor as a concept that is disposed with the aforementioned centroid.

13 and 14 are diagrams for explaining a mutual factor (or cross factor) according to an embodiment.

13 shows the waveform when the principal component is obtained for the first mode of the bitstream file converted to the 0th mode, and FIG. 14 is the bitstream file converted to the 7th mode in the fourth mode. The waveform is shown when the principal component is obtained. In FIGS. 13 and 14 , the blue graph and the red graph represent graphs of male and female voices, respectively.

13 and 14 , when the vector average is obtained in an inappropriate mode, even correlation information remaining in the codec bitstream is lost and only values close to 0.5 are observed in the graph. That is, through the graphs of FIGS. 13 and 14 , it can only be interpreted that the amplitude of the voice signal included in the AMR file is very weak, or that the AMR file contains no information available at all. No, these principal components become cross-factors.

15 is a diagram illustrating a principal component matrix for each mode generated by matching an input mode in which target data is encoded and a check mode comparing the same according to an embodiment of the present invention. The horizontal axis is an input mode in which the target data is encoded, and includes modes 0 to 7, and the vertical axis is a check mode for extracting a principal component with respect to the target data input mode. Includes 7 modes. That is, the principal component matrix of FIG. 15 may have the form of an 8×8 matrix including both eigen factors and mutual factors for all modes of the AMR format, and (i, j)(i, j is a natural number less than or equal to 8).

Referring to FIG. 15 , a graph located at a diagonal position having a coordinate value of (m, m) (m is a natural number less than or equal to 8) is a principal component graph of the coincidence mode extracted with the same check mode as the input mode, and is based on the eigen factor. corresponds to On the other hand, as the positions other than the diagonal positions, the graph of positions having coordinate values of (m, n) (m, n are natural numbers less than or equal to 8 and have different values) is a check mode different from the input mode. As the principal component graph of the extracted discordance mode, it corresponds to the above-mentioned mutual factor. It can be seen that the intrinsic factor located at the diagonal position has a distinct pattern in which the amplitude change is visibly compared with the mutual factor of the other positions. With the exception of some, it can be seen that the signal magnitude is weak or the amplitude change is insignificant when compared to the intrinsic factor.

In the present disclosure, in order to determine whether the target speech data is AMR speech data, estimation using the conditional probability of the MSE distribution is performed. In the present disclosure, as for the individual distributions of the conditional probabilities, it is assumed that the MSE distribution is a Gaussian distribution, and it is assumed that the overall distribution of the MSE is a linear combination of Gaussian distributions having different properties.

The MSE acquisition unit 310 for each mode may receive the centroid from the target data processing unit 100 and the intrinsic factor from the reference data processing unit 200 to obtain the MSE distribution for each mode. Hereinafter, a method for obtaining the MSE distribution for each mode will be described in detail.

Hereinafter, a codeword generated (defined) based on the intrinsic factor extracted from the main component of the reference data is referred to as c _i , and the centroid, which is a unique factor extracted from the main component of the target data to be estimated, is referred to as d _i , MSE The value may be defined as in [Equation 2] below.

In order to discriminate the AMR voice data according to an embodiment of the present invention, the MSE distribution for every position in the matrix of FIG. 15 is required. In other words, with respect to the input mode of the target data, the MSE distribution for all principal component graphs extracted based on the different check modes as well as the same check mode is required. Hereinafter, a case in which the input mode and the check mode are the same may be referred to as a 'match mode', and a case in which the input mode and the check mode are different from each other may be referred to as a 'mismatch mode', and the matching mode is (m, m in the matrix of FIG. ) and the mismatch mode may have a coordinate value of (m, n) in the matrix of FIG. 15 . In addition, the eigen factor may be obtained through the principal component of the coincident mode, and the mutual factor may be obtained through the principal component of the mismatch mode.

In this case, the conditional probability distribution of the MSE in concordance mode is defined as N ₁ (μ ₁ , σ ₁ ) (μ ₁ is the mean of the consensus mode, σ ₁ is the standard deviation of the consensus mode), and the conditional probability distribution of the MSE in the mismatch mode is defined as is defined as N ₂ (μ ₂ , σ ₂ ), where μ ₂ is the mean of the inconsistent mode, and σ ₂ is the standard deviation of the inconsistent mode.

For each mode, it is possible to obtain reference data based on male voice data, extract a unique factor therefrom, and obtain a codeword. In parallel, target data may be obtained based on the female voice data for each mode, and centroids may be extracted therefrom. Using the codewords and centroids obtained in this way, 10 matching mode data (c _i , d _i ) and 70 non-matching mode data (c _i , d _i ) are substituted into Equation 2 to obtain the MSE value. It is possible to obtain a total of 80 constant values for . By reversing the male and female data, a codeword is obtained based on the female voice data, and the centroid is extracted based on the male voice data, and a total of 80 MSE constant values can also be obtained from this.

In this case, the number of data of 10 and 70, respectively, in the coincident mode and the non-matching mode is only an example, and it is sufficient if the data differs by 7 times from each other or a similar number of data is secured for each coordinate of the matrix.

As described above, graphs in which the MSE distribution, which is a probability density function of the MSE values obtained for each mode, is calculated as an average value is shown in FIGS. 16 and 17 to be described later.

16 is a graph showing mean square error (MSE) distribution according to an embodiment of the present invention with respect to mode 0 to mode 3, and FIG. 17 is an embodiment of the present invention with respect to mode 4 to mode 7 It is a graph showing the MSE distribution according to the example. The graph shown by the red line is the first MSE distribution (N ₁ ) as the MSE distribution in code match mode, and the graph shown by the blue line is the MSE distribution in the code mismatch mode as the second MSE distribution (N ₂ ). ) is indicated. The first MSE distribution may be a distribution calculated as an average of the above-described 10 pieces of data, and the second MSE distribution may also be a distribution calculated as an average of the above-described 70 pieces of data.

16 and 17 , it can be seen that the first MSE distribution (N ₁ ) has a narrow amplitude and a small dispersion, whereas the second MSE distribution (N ₂ ) has a wide amplitude and a large dispersion (σ). ₁ <σ ₂ ). On the other hand, it can be seen that the average value of the first MSE distribution (N ₁ ) is generally smaller than that of the second MSE distribution (N ₂ ) (μ ₁ <μ ₂ ).

It can be determined whether the target data is AMR data through the experimental estimation of the conditional probability distribution of the MSE as described above. In this case, for the determination, it is necessary to obtain a threshold value using the relationship between the first MSE distribution and the second MSE distribution.

The threshold value acquisition unit 320 may acquire the threshold value of the present invention by a method to be described later. Since the MSE distribution is assumed to be a Gaussian distribution (normal distribution) in the present disclosure, the probability density function f(μ,σ) of the Gaussian distribution may be expressed by the following [Equation 3].

At this time, as an equation for obtaining the threshold value, a point (x) at which the aforementioned first MSE distribution (N ₁ ) (probability density function of) and the second MSE distribution (N ₂ ) (probability density function of) have the same value In order to obtain a threshold value, the following [Equation 4] can be derived. The point of the threshold value x may be a right intersection point of N ₁ and N ₂ in the graphs of FIGS. 16 and 17 . In other words, a position where the two MSE distributions (N ₁ , N ₂ ) are identical to each other may be used as a threshold value.

According to Equation 4, the threshold value x may be calculated as in Equation 5 below.

Experimental data (σ ₁ , for each mode described above based on Equation 5) σ ₂ , μ ₁ , μ ₂ ) is used to obtain the threshold values for each mode, for example, the data shown in Table 3 below can be obtained. The threshold data according to [Table 3] below is merely an example and may be a value different from the value of Table 3 below, or a value near the value specified in Table 3.

mode Threshold (x) 0 0.001243 One 0.000789 2 0.000839 3 0.000719 4 0.000297 5 0.000528 6 0.00019 7 0.000349

Thereafter, the comparison determination unit 330 compares the minimum value of the MSE distribution of the target data and the threshold value of the input mode in which the target data is compressed AMR mode among the threshold values for each mode to determine whether the target data is AMR voice data. can be judged. More specifically, the comparison determination unit 330 may extract a minimum value of the MSE distribution of the target data and compare the minimum value with the threshold value. That is, a maximum likelihood method of estimating the direction in which the MSE distribution is minimized may be used.

The comparison determination unit 330 may determine that the target data is AMR voice data when the minimum value of the MSE distribution is less than the threshold value. Conversely, when the minimum value of the MSE distribution is greater than or equal to the threshold value, it may be determined that the target data includes non-AMR speech data compressed in a format other than the AMR format. In other words, this case may include not only a case in which the target data includes only non-AMR voice data, but also a case in which the target data includes both AMR voice data and non-AMR voice data.

For example, assuming that the target data is compressed in the first mode, the input mode is the first mode, and the threshold value of the first mode is 0.000789 according to Table 3 above. When the minimum value of the MSE distribution of the target data is x _min , if the minimum value (x _min ) is less than 0.000789 (x _min < 0.000789), it belongs to the first MSE distribution of the matching mode indicated by the red line in the graph of FIG. 16 . Therefore, it can be determined that the target data is AMR data.

18 is a flowchart illustrating a method for determining AMR voice data according to an embodiment of the present invention. Since FIG. 18 can be performed by the apparatus of FIGS. 1 and 2 , it will be described with reference to FIGS. 1 and 2 , and overlapping descriptions may be omitted or simplified.

First, the target data processing unit 100 may extract a centroid from the compressed target data in the AMR format (S100).

The reference data processing unit 200 may similarly extract a reference intrinsic factor, which is a unique factor of the reference data, from the reference data compressed in the AMR format ( S200 ). A codeword may be generated (defined) based on the reference intrinsic factor.

Although the figure shows that step S200 is performed after step S100, steps S100 and S200 may be performed in parallel.

Thereafter, the comparison unit 300 may determine whether the target data is AMR voice data by comparing the target data and the reference data using a Maximum Likelihood (ML) method ( S300 ). The comparison unit 300 may determine using the MSE distribution, which is a probability density function of a mean squared error (MSE) between the centroid and the reference eigen factor. The comparison determination step ( S300 ) may include steps to be described later.

First, the MSE acquisition unit 310 for each mode may acquire the MSE distribution for each mode based on the centroid and the reference intrinsic factor ( S310 ).

Thereafter, the threshold value acquisition unit 320 may acquire a threshold value for each mode based on the MSE distribution for each mode ( S320 ). In this case, the threshold value is the first MSE distribution (N ₁ ) obtained from the principal component of the concordance mode and the second MSE distribution (N ₂ ) obtained from the principal component of the mismatch mode according to the above-mentioned [Equation 4] and [Equation 5]. ) can be obtained as a point having the same value. More specifically, the first MSE distribution (N ₁ ) is obtained using the centroid and codewords (c _i , d _i ) of concordance mode, and the centroid and codewords (c _i , d _i ) of discordant mode are obtained. A second MSE distribution (N ₂ ) may be obtained using .

Thereafter, the comparison determination unit 330 may determine whether the target data is AMR voice data by comparing the MSE distribution value of the target data with the threshold value of the input mode among the threshold values for each mode.

In this case, the comparison determination unit 330 may first obtain the minimum value of the MSE distribution of the target data and determine whether the minimum value is smaller than the threshold value ( S330 ). When the minimum value of the MSE is smaller than the threshold value (S330-Y), it may be determined that the target data is AMR voice data (S340). Conversely, when the minimum value of the MSE is greater than or equal to the threshold value (S330-N), the target data may be determined to be non-AMR voice data rather than AMR voice data (S350). In this case, the non-AMR voice data may include not only a case in which the target data includes only non-AMR voice data, but also a case in which the target data includes both AMR voice data and non-AMR voice data.

The various embodiments described herein are illustrative, and do not need to be performed independently of each other. The embodiments described in this specification may be implemented in combination with each other.

The various embodiments described above may be implemented in the form of a computer program that can be executed through various components on a computer, and such a computer program may be recorded in a computer-readable medium. In this case, the medium may be to continuously store the program executable by the computer, or to temporarily store the program for execution or download. In addition, the medium may be various recording means or storage means in the form of a single or several hardware combined, it is not limited to a medium directly connected to any computer system, and may exist distributed on a network. Examples of the medium include a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk, and those configured to store program instructions, including ROM, RAM, flash memory, and the like. In addition, examples of other media include recording media or storage media managed by an app store that distributes applications, sites that supply or distribute various other software, and servers.

In the above, preferred embodiments of the present invention have been illustrated and described, but the present invention is not limited to the specific embodiments described above, and it is common in the technical field to which the present invention pertains without departing from the gist of the present invention as claimed in the claims. Various modifications are possible by those having the knowledge of, of course, and these modifications should not be individually understood from the technical spirit or perspective of the present invention.

Therefore, the spirit of the present invention should not be limited to the above-described embodiments, and not only the claims described below, but also all scopes equivalent to or changed from the claims described below are of the spirit of the present invention. would be said to belong to the category.

10: AMR voice data discrimination device
100: target data processing unit
200: reference data processing unit
110, 210: main component extraction unit
120: centroid extraction unit
220: codeword extraction unit
300: comparison unit
310: MSE acquisition unit for each mode
320: threshold value acquisition unit
330: comparison determination unit

Claims (15)

A target data processing step of extracting a centroid, which is a unique factor of the target data, from the target data compressed in the AMR format of one of the 0th mode to the 7th mode;
a reference data processing step of extracting a reference intrinsic factor, which is an intrinsic factor of the reference data, from the reference data compressed in the AMR format; and
a comparison determination step of determining whether the target data is AMR data compressed in an AMR format by comparing the centroid and the reference intrinsic factor using a Maximum Likelihood (ML) method;
including,
The comparison and determination step is determined using an MSE distribution that is a probability density function of a mean squared error (MSE) between the centroid and the reference eigen factor. According to claim 1,
The target data processing step is,
extracting a principal component from the target data; and
extracting the centroid for a partial region of the main component from the main component;
including,
The extracting of the centroid may include dividing the main component into a plurality of subframes including at least one or more bits, and the centroid based on bit information included in some subframes of the plurality of subframes. AMR voice data discrimination method comprising the step of specifying. 3. The method of claim 2,
An input mode in which the target data is a compressed AMR mode, and a check mode for providing the number of bits for extracting the principal component with respect to the input mode include the 0th mode to the 7th mode,
The step of extracting the main component,
extracting a principal component of the matching mode extracted based on the same check mode as the input mode with respect to the input mode; and
and extracting a cross factor, which is a main component of a mismatch mode extracted based on a check mode different from the input mode, with respect to the input mode. 3. The method of claim 2,
The step of extracting the main component,
a data receiving step of receiving a bitstream of target data compressed in one of the 0th mode to the 7th mode;
a frame separation step of dividing the bitstream into frames having a predetermined first number of bits;
a bit array generating step of generating a bit array by integrating the separated frames; and
specifying the principal component of the bitstream based on the bit array and the total number of frames;
including,
The centroid is a method for determining AMR voice data in which the centroid is extracted based on some of the frames. 3. The method of claim 2,
The reference data processing step is,
extracting a reference principal component from the reference data; and
extracting the reference eigen factor for a partial region of the reference principal component from the reference principal component;
including,
The extracting of the reference eigen factor may include dividing the reference principal component into a plurality of subframes including at least one bit, and based on bit information included in some subframes of the plurality of subframes, the A method for determining AMR speech data, comprising the step of specifying a reference intrinsic factor. 4. The method of claim 3,
The comparison determination step is
obtaining an MSE distribution for each mode based on the centroid and the reference intrinsic factor;
obtaining a threshold value for each mode based on the MSE distribution for each mode; and
comparing the minimum value of the MSE distribution of the target data with the threshold value of the input mode among the threshold values for each mode, and determining that the target data is AMR voice data when the minimum value is less than the threshold value;
Including, AMR voice data discrimination method. 7. The method of claim 6,
The step of obtaining the threshold value is
and acquiring, as a threshold, a point at which the first MSE distribution of the principal component of the inconsistent mode has the same value as the second MSE distribution of the principal component of the inconsistent mode. A computer-readable recording medium storing a program for executing the method according to any one of claims 1 to 7. a target data processing unit for extracting a centroid, which is a unique factor of the target data, from the target data compressed in one of the 0th mode to the 7th mode;
a reference data processing unit for extracting a reference intrinsic factor, which is an intrinsic factor of the reference data, from the reference data compressed in the AMR format; and
A comparison unit for determining whether the target data is AMR data compressed in an AMR format by comparing the centroid and the reference intrinsic factor using a Maximum Likelihood (ML) method;
The comparison unit determines using the MSE distribution, which is a probability density function of a mean squared error (MSE) between the centroid and the reference eigen factor, to determine the AMR speech data. 10. The method of claim 9,
The target data processing unit,
a principal component extracting unit for extracting a principal component from the target data; and
a centroid extraction unit for extracting the centroid for a partial region of the main component from the main component;
including,
The centroid extraction unit divides the main component into a plurality of subframes including at least one or more bits, and specifies the centroid based on bit information included in some subframes of the plurality of subframes, AMR Voice data discrimination device. 11. The method of claim 10,
An input mode in which the target data is a compressed AMR mode, and a check mode for providing the number of bits for extracting the principal component with respect to the input mode include the 0th mode to the 7th mode,
The main component extraction unit,
extracting the principal components of the matching mode extracted based on the same check mode as the input mode with respect to the input mode;
An apparatus for determining AMR voice data for extracting a cross factor, which is a main component of a discrepancy mode extracted based on a check mode different from the input mode, with respect to the input mode. 11. The method of claim 10,
The main component extraction unit,
a data receiver configured to receive a bitstream of target data compressed in an AMR format among the 0th mode to the 7th mode;
a frame dividing unit dividing the bitstream into frames having a predetermined first number of bits;
a bit array generator for generating a bit array by integrating the separated frames; and
a principal component specifying unit for specifying the main component of the bitstream based on the bit array and the total number of frames;
including,
The centroid is extracted based on some of the frames, AMR voice data discrimination device. 11. The method of claim 10,
The reference data processing unit,
a reference principal component extracting unit for extracting a reference principal component from the reference data; and
Including;
The reference intrinsic factor extracting unit separates the reference main component into a plurality of subframes including at least one bit, and specifies the reference intrinsic factor based on bit information included in some subframes of the plurality of subframes. AMR voice data discrimination device. 12. The method of claim 11,
The comparison unit,
an MSE acquisition unit for each mode for acquiring an MSE distribution for each mode based on the centroid and the reference intrinsic factor;
a threshold value acquisition unit for acquiring a threshold value for each mode based on the MSE distribution for each mode; and
a comparison determining unit that compares the minimum value of the MSE distribution of the target data and the threshold value of the input mode among the threshold values for each mode, and determines that the target data is AMR voice data when the minimum value is less than the threshold value;
Including, AMR voice data discrimination device. 15. The method of claim 14,
The threshold value acquisition unit,
and acquiring a point at which the first MSE distribution of the principal component of the coincident mode has the same value as the second MSE distribution of the principal component of the inconsistent mode as a threshold value.

Images