KR20110115984A - Method, apparatus, and program containing medium for measurement of audio quality - Google Patents

Abstract

Comparing the reference signal with the test signal to generate one or more model output variables including a variable representing an envelope difference time difference distortion ( EITDDist ); and combining the one or more model output variables to a value corresponding to sound quality. The sound quality measurement method comprising the step of outputting is disclosed.

Description

Sound quality measurement method, sound quality measurement device, sound quality measurement program recording medium {Method, apparatus, and program containing medium for measurement of audio quality}

The present invention relates to a sound quality measurement method, a sound quality measurement device, a sound quality measurement program recording medium, and more particularly, to an objective sound quality measurement technology.

Objective sound quality evaluation is one of the important applications in psychoacoustic field. Such objective sound quality evaluation is widely used for quality evaluation of compression technique of mono and stereo sound sources.

Recommendations have been adopted on how to assess the sound quality of single channel audio signal compression codecs in the ITU Radiocommunication Sector (ITU-R) of the International Telecommunication Union (ITU-R Recommendation BS.1387-). 1, "Method for objective measurements of perceived audio quality", International Telecommunication Union, Geneva, Switzerland, 1998). This recommendation may not be suitable for assessing the sound quality of medium and low performance audio compression codecs and multichannel audio compression codecs. In addition, objective measurement of this Recommendation focuses on applications that are subjectively assessed by applying ITU-R BS.1116-1 (Methods for the Subjective Assessment of Small Impairments in Audio Systems Including Multichannel Sound Systems).

On the other hand, the multi-channel audio compression codec is being actively discussed in the MPEG standardization group (ISO / IEC / JTC1 / SC29 / WGll), and codecs developed by various organizations have been announced. The sound quality evaluation of these codecs is based on the 'MUSHRA' technique. Based on the listening subjective assessment technique (ITU-R Recommendation BS. 1534-1, "Method for the Subjective Assessment of Intermediate Sound Quality (MUSHRA)", International Telecommunication Union, Geneva, Switzerland, 2001). The results of listening evaluations for a number of codecs performed using the software have been published (ISO / IEC JTC1 / SC29 / WGll (MPEG), N7138, "Report on MPEG Spatial Audio Coding RMO Listening Tests," ISO / IEC JTC1 / SC29 /). WG11 (MPEG), N7139, "Spatial Audio Coding RM0 listening test data").

However, in evaluating the sound quality of the multi-channel audio compression codec, the sound quality evaluation method in which the listener directly listens and evaluates the sound quality and undergoes statistical processing thereof is subjective. There is a need for a method to omit evaluation and statistical processing, to perform sound quality evaluation through a consistent measurement of sound quality, or to predict a sound quality evaluation result.

As the multi-channel system becomes more common and the multi-channel compression technique is developed, the need for objective sound quality evaluation for the multi-channel audio signal is emerging.

The present invention develops an evaluation factor for objective evaluation of a multi-channel audio compression codec, and uses the evaluation factor to describe a method, an apparatus for evaluating an audio compression codec, and a technique for recording a program for performing the method. The purpose is to derive. However, the scope of the present invention is not limited by the above-mentioned object.

In order to evaluate the objective sound quality of the multi-channel audio signal, the existing ITU-R Rec. BS. The sound quality prediction model of 1387-1 can be extended to a multichannel signal. This extension model is based on ITU-R Rec. BS. In addition to the 10 timbre components used in 1387-1 , at least three more spatial components can be used, such as bilateral time-distortion ( ITDDist ), bilateral magnitude-distortion ( ILDDist ), and bilateral correlation distortion ( ICACCist ). .

In particular, a positive time difference distortion ( ITDDist ) may be used as an element for predicting an error in a sound image position. As the positive time difference distortion, not only the positive time difference distortion in the low frequency band where the positive phase difference ( IPD ) is apparent, but also the positive time difference distortion for the envelope of the high frequency band can be used.

In general, when humans recognize the location of low and high frequency sound sources, the brain undergoes different processes. A positive time difference is used for recognizing the location of a low frequency sound source. Excitation patterns generated in the basal membrane by low-frequency sound stimulation are transmitted to the Medial Superior Olive (MSO), and the transmitted signals are processed by coincidence detection neurons. Yang calculates the time difference, and humans use it to recognize the location of the sound source.

On the other hand, for the high frequency sound source, the stimulation pattern of the basement membrane is transferred to the Lateral Superior Olive (LSO), which causes different size electric signals to be generated at both sides of the upper olive nucleus. By doing this, the human being perceives the position of the high frequency sound source. However, in addition to the magnitude difference, the envelope information of the high frequency sound source may also be used for the sound image positioning of the high frequency sound source. In particular, neurons present in the lateral rib nucleus have a sensitivity to high frequency transposed tones. In addition, the neural firing probability of the auditory fiber (ANF) for high frequency transposed tones is similar to that of the nerve fiber for low frequency sound sources. In addition, the sensitivity of the high frequency envelope to the time difference is similar to the sensitivity of the low frequency sound source to the time difference. In view of this point, it can be judged that the amount of high frequency region envelope has a time difference, and the amount of low frequency region has a large influence on the sound position along with the time difference and the amount of high frequency region.

In the technique according to an aspect of the present invention, the amount of high frequency envelope uses an EITDDist ( Envelope Interaural Time Difference distortion) as an objective evaluation factor of a multichannel audio signal.

According to an aspect of the present invention, there is provided a sound quality measuring method comprising: generating at least one model output variable (MOV) including a variable representing an envelope positive time difference distortion by comparing a reference signal and a test signal, And outputting a value corresponding to audio quality by combining the one or more model output variables.

In this case, the output may include inputting the one or more model output variables to an artificial neural network to generate a value corresponding to the sound quality.

According to another aspect of the present invention, a computer-readable medium includes: generating, by a computer, one or more model output variables including a variable representing an envelope positive time difference distortion by comparing a reference signal and a test signal, and A program for executing the step of outputting a value corresponding to sound quality by combining the one or more model output variables is recorded.

According to another aspect of the present invention, a computer-readable medium may include a code for changing a program for outputting a value corresponding to sound quality by combining one or more model output variables generated by comparing a reference signal and a test signal. As a computer-readable medium, the code is adapted to modify the program such that a variable representing an envelope positive time difference distortion obtained by comparing the reference signal with the test signal is included in the one or more model output variables.

An apparatus for measuring sound quality according to another aspect of the present invention includes a variable indicating an envelope interaural time difference ( EITDDist ) by comparing a reference signal and a signal under test. Model output variable generating means for generating one or more model output variables (MOV), and output means for outputting a value corresponding to audio quality by combining the one or more model output variables. It includes.

In this case, the generating means and the output means, by comparing the reference signal and the test signal to generate at least one model output variable including a variable representing an envelope positive time difference distortion, and the at least one model It may be part of a processing apparatus capable of driving a program for executing the step of outputting a value corresponding to sound quality by combining output variables.

로 주어지며,

는 상기 기준 신호와 상기 테스트 신호의 n번째 시간 프레임의 k번째 주파수 밴드를 비교하여 생성한 포락선 양이시간차 왜곡을 나타낼 수 있다. In various aspects of the present invention described above, EITDDist, which is a variable representing the time difference distortion of the envelope,

Given by

May represent an envelope positive time difference distortion generated by comparing the reference signal with a k-th frequency band of an n-th time frame of the test signal.

는

로 주어지고, 상기

는 상기 기준 신호와 상기 테스트 신호의 n번째 시간 프레임의 k번째 주파수 밴드에서의 포락선 양이시간차(EITD, Envelope Interaural Time Difference)의 차이값을 나타내며, 상기

는 상기 테스트 신호의 n번째 시간 프레임의 k번째 주파수 밴드에서의 포락선 양이 상관계수(EIACC, Envelope InteAural Cross-correlation Coefficient)의 비선형 변환값이고, 상기

는 상기 기준 신호의 n번째 시간 프레임의 k번째 주파수 밴드에서의 포락선 양이 상관계수(EIACC, Envelope InteAural Cross-correlation Coefficient)의 비선형 변환값일 수 있다.In addition, in various aspects of the present invention described above,

Is

Given by

Denotes a difference value between an envelope difference time difference ( EITD ) in a k-th frequency band of an n-th time frame of the reference signal and the test signal.

Is the nonlinear transform value of the envelope coefficient in the k-th frequency band of the n-th time frame of the test signal ( EIACC , Envelope InteAural Cross-correlation Coefficient).

The envelope amount in the k th frequency band of the n th time frame of the reference signal may be a nonlinear transform value of the correlation coefficient ( EIACC , Envelope InteAural Cross-correlation Coefficient).

In addition, in various aspects of the present invention described above, the reference signal is generated from a multichannel audio signal, and the test signal passes the multichannel audio signal to a device under test to measure the sound quality. It may be generated by.

Also, in various aspects of the present invention described above, at least one of the one or more model output variables may be generated by comparing an excitation pattern of the reference signal and the test signal.

In addition, in various aspects of the present invention described above, the variable representing the envelope positive time difference distortion may be generated by passing the reference signal and the test signal through a filter bank.

According to the present invention, the performance of the objective evaluation model of the multi-channel audio codec can be improved by using a variable relating to the envelope difference time difference distortion. The scope of the present invention is not limited by the above-mentioned effects.

1 is an exemplary configuration diagram of a multi-channel audio reproduction system recommended by ITU-R that can be applied to an embodiment of the present invention.
2 is a block diagram of a sound quality evaluation apparatus of a multi-channel audio compression codec according to an embodiment of the present invention.
3 graphically illustrates ten sound transmission paths.
4 is an operation explanatory diagram of a preprocessor of the sound quality evaluation apparatus according to an embodiment of the present invention.
5 is a flowchart illustrating a sound quality estimation method of a multi-channel audio compression codec according to an embodiment of the present invention.
6 is a flowchart for calculating an ILD distortion, and FIG. 7 is a flowchart for calculating an EITD distortion according to an embodiment of the present invention.
8 shows an example of envelope extraction.
9 illustrates the flow of calculating EITD distortion in accordance with FIG. 7 in more detail.

The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, whereby those skilled in the art may easily implement the technical idea of the present invention. There will be. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail. Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.

ITU-R Recommendation BS.1116-1, "Methods for the Subjective Assessment of Small Impairments in Audio Systems Including Multichannel Sound Systems", ITU-R Recommendation BS. 1387-1, "Method for objective measurements of perceived audio quality", International Telecommunication Union, Geneva, Switzerland, 1998, and ITU-R Recommendation BS. 1534-1, "Method for the Subjective Assessment of Intermediate Sound Quality (MUSHRA)", International Telecommunication Union, Geneva, Switzerland, 2001 and ISO / IEC JTC1 / SC29 / WGll (MPEG), N7138, "Report on MPEG Spatial Audio Coding RMO Listening Tests, "ISO / IEC JTC1 / SC29 / WG11 (MPEG), N7139," Spatial Audio Coding RM0 listening test data ", are incorporated herein by reference.

In general, multichannel audio includes front speakers (LF (Left Front), RF (Right Front), center speaker (C (Center)), bass channel (LEF (Low Frequency Effect)), rear speakers (LS (Left Surround)). ), 6 channels (or 5.1 channels) of RS (Right Surround), and since the double bass channel (LEF) is not actually used as a low-frequency effect channel, in the embodiment of the present invention, the front speaker (LF) Only five channels, RF), center speaker (C) and rear speaker (LS, RS) will be used.

1 is an exemplary configuration diagram of a multi-channel audio reproduction system recommended by ITU-R, which may be applied to an embodiment of the present invention.

As shown in Fig. 1, in the multi-channel audio reproduction system recommended by the ITU-R, five-channel speakers are arranged on one circumference around the listener 10, and the left and right front speakers (L, R) and the listener are located. 10 forms an equilateral triangle, and the distance between the front center speaker C and the listener 10 is equidistant from the left and right front speakers L and R, and the left and right rear speakers LS and RS are zero degrees in front of the front. It can be located on the concentric circle of 100 degrees to 120 degrees.

The reason for this standard arrangement recommended by the ITU-R is that most sources are edited and recorded to meet this placement standard, so that the intended sound quality (best sound quality) can be obtained by following this standard arrangement. .

In an embodiment of the present invention, a 5-channel speaker (L, R, C) using a microphone that simulates the listener 10 of the standard multichannel audio reproduction system recommended by ITU-R and the human body (head and torso) is used. By measuring the impact response of the multi-channel audio signal from (LS, RS), it can be replaced by the sound quality evaluation device of the multi-channel audio compression codec for evaluating sound quality.

2 is a block diagram of a sound quality evaluation apparatus of a multi-channel audio compression codec according to an embodiment of the present invention.

)를 생성하기 위한 전처리부(11)와, 전처리부(11)에 의해 생성된 양이 입력 신호(

)의 양이 상관 정도 왜곡(IACCDist), 양이 크기 차이 왜곡(ILDDist), 및 고주파 포락선 양이시간차 왜곡(EITDDist, Envelope Interaural Time Difference distortion)을 포함하는 모형출력변수들(MOV, Model Output Variables)을 산출하기 위한 출력변수 계산부(12)와, 출력변수 계산부(12)로부터 위의 모형출력변수들을 입력받아, 이를 바탕으로 음질의 등급을 출력하기 위한 인공신경망회로부(13)를 포함할 수 있다.As shown in Fig. 2, the sound quality evaluation device 10 of the multi-channel audio compression codec is input from each channel (L, R, C, LS, RS) of the standard multi-channel audio reproduction system recommended by ITU-R. Based on a multichannel audio signal

) And the amount generated by the preprocessor 11 generates an input signal (

Model Output Variables (MOV), including positive correlation distortion ( IACCDist ), positive magnitude difference distortion ( ILDDist ), and high-frequency envelope positive time difference distortion ( EITDDist ). It may include an output neural network unit 13 for calculating the output variable calculation unit 12, and the above model output variables from the output variable calculation unit 12, and outputs the grade of sound quality based on this. have.

Here, the degree of bilateral correlation ( IACC ) represents the degree of correlation of the signal input to both ears (sheep), and the amount of interaural level difference (ILD) is input to both ears (sheep). It can represent the energy ratio of the signal to be. In addition, the high frequency envelope difference time difference EITD may represent a difference in time when an envelope of an audio signal of a high frequency band is input to both ears.

Hereinafter, the operation of the components of the apparatus for evaluating sound quality of a multi-channel audio compression codec according to the present invention will be described. Five channels of a sound source encoded and decoded by the multi-channel audio compression codec to be evaluated are LF _test , RF _test , C _test , LS _test and RS _test are indicated, and the original five channels are represented by LF _ref , RF _ref , C _ref , LS _ref , and RS _ref .

In this document, LF _test , RF _test , LF _ref , and RF _ref may be referred to as L _test , R _test , L _ref , and R _ref , respectively.

)를 산출할 수 있다.First _, a total of 10 signals of LF _test , RF _test , C _test , LS _test , RS _{test ,} LF _ref , RF _ref , C _ref , LS _ref , and RS _ref are input to the preprocessor 11, and the preprocessor 11 Is a function of the impact response of the sound transmission path measured using a microphone that simulates the human body (head and torso) of the standard multichannel audio playback system recommended by the ITU-R.

) Can be calculated.

)의 양이 상관 정도 왜곡(IACCDist), 양이 크기 차이 왜곡(ILDDist), 고주파 포락선 양이시간차 왜곡(EITDDist)을 포함하는 변수들을 출력변수들로 산출하여, 이 출력변수들을 인공신경망회로부(13)에 입력하고, 인공신경망회로부(13)는 출력변수 계산부(12)로부터 입력된 이 출력변수들을 바탕으로 객관적인 음질 등급(ODG, Objective Difference Grade)을 출력할 수 있다. At this time, the total sound transmission path is ten, which can be represented by a graph as shown in FIG. The output variable calculator 12 has two input signals inputted from the preprocessor 11.

) Are computed as output variables including the amount of correlation correlation distortion ( IACCDist ), the magnitude difference distortion ( ILDDist ), and the high frequency envelope quantum time difference distortion ( EITDDist ), and then output these output variables to the artificial neural network unit 13. ), And the artificial neural network unit 13 may output an objective sound grade (ODG) based on the output variables input from the output variable calculator 12.

)의 양이 크기 차이 왜곡(ILDDist)을 산출할 수 있다. 압축되지 않은 본래 오디오 신호의 양이 크기 차이(ILD)를 ILD_ref라 하고, 멀티채널 오디오 압축 코덱에 의해 부호화 및 복호화된 오디오 신호의 양이 크기 차이(ILD)를 ILD_test라 한다. 양이 상관정도(IACC)에 대해서도 이와 같이 명명할 수 있다. 양이 상관 정도(IACC) 및 양이 크기 차이(ILD)는 44100Hz의 주파수로 샘플링되었을 때 2048개의 샘플을 50%씩 중첩하며 진행하는 시간 프레임에서 24개의 청각 임계대역에 대해 각각 계산될 수 있다. 이중 n번째 프레임의 k번째 대역에서 양이 크기 차이 왜곡(ILDDist)은 ILDDist[k,n]으로 표시할 수 있다.Here, the output variable calculator 12 uses two equations inputted from the preprocessor 11 using equations (1) and (2).

) May yield magnitude difference distortion ( ILDDist ). La this size difference (ILD) the amount of the original audio signal that is not compressed ILD _ref, and the amount of the audio signal coding and decoding by the multi-channel audio compression codec, and the size difference (ILD) ILD _test la. The same can be said for quantity correlation ( IACC ). The amount of positive correlation ( IACC ) and the amount of positive difference ( ILD ) can be calculated for 24 auditory critical bands in a time frame that progresses by 50% overlapping 2048 samples when sampled at a frequency of 44100 Hz. The positive magnitude difference distortion ( ILDDist ) in the k-th band of the n-th frame may be represented by ILDDist [k, n].

[Equation 1]

In this case, ILDDist is a magnitude difference distortion ( ILDDist ), and w [k] is a weight function determined according to a critical band range, and may reflect auditory sensitivity with respect to magnitude difference ( LDD ).

Meanwhile, the amount of all auditory bands in the nth time frame may be averaged as in Equation 2 to obtain magnitude difference distortion ( ILDDist ).

[Equation 2]

As shown in equation (2), the former by averaging over the time frames, and the amount of the multi-channel compression codec can calculate the size difference distortion (ILDDist), can be calculated the amount also applies to any degree (IACC). At this time, the positive correlation distortion ( IACCDist ) is represented by ICCDist , and the positive magnitude difference distortion ( ILDDist ) and the positive correlation distortion ( ICCDist ) are the result of the sound quality evaluation (subjective evaluation) of the multi-channel audio compression codec by the listener. Since it has a high correlation, the output variable calculation unit 12 may regard this as an output variable, and input this value and other possible output variables to the artificial neural network unit 13 to output an objectivity and a consistent sound quality grade. have.

A detailed method of calculating the high frequency envelope lag distortion EITDDist in the output variable calculator 12 will be described with reference to FIG. 8.

4 is an operation explanatory diagram of an embodiment of a preprocessor of the sound quality evaluation apparatus according to the present invention.

)를 산출할 수 있다.As shown in FIG. 4, the preprocessing unit 11 of the sound quality evaluation apparatus 10 measures the amount of the human body (head and torso) of the standard multichannel audio reproduction system recommended by the ITU-R using a microphone. Transmits the shock response of the sound transmission path, sums them up, and adds the positive input signal (

) Can be calculated.

5 is a flowchart illustrating a sound quality estimation method of a multi-channel audio compression codec according to the present invention.

)를 산출할 수 있다(S501).First, the preprocessing unit 11 of the sound quality evaluation apparatus 10 of the multi-channel audio compression codec transfers the shock response of the sound source encoded and decoded by the multi-channel audio compression codec and the original sound source, sums them, and inputs a positive input. signal(

) Can be calculated (S501).

)의 양이 상관 정도 왜곡(IACCDist), 양이 크기 차이 왜곡(ILDDist), 고주파 포락선 양이시간차 왜곡(EITDDist)을 포함하는 출력변수들을 산출하여(S502), 이 출력변수들을 인공신경망회로부(13)에 입력할 수 있다(S503).Thereafter, the output variable calculator 12 receives two input signals from the preprocessor 11.

) Are calculated by calculating the output variables including the correlation degree distortion ( IACCDist ), the difference magnitude difference distortion ( ILDDist ), and the high frequency envelope positive time difference distortion ( EITDDist ) (S502). ) Can be entered (S503).

In addition, the artificial neural network unit 13 may output a sound quality grade based on the output variables input from the output variable calculator 12 (S504).

)의 포락선(Envelope)의 양이 시간 차이(EITD, Envelope Interaural Difference)가 더 출력될 수 있다. 포락선의 양이 시간 차이(EITD)는 인공신경망회로부(13)에 더 입력될 수 있다.
In the output variable calculator 12 of FIG. 2, the amount generated by the preprocessor 11 is determined by the input signal (

The time difference EITD (Envelope Interaural Difference) may be further output. The amount of envelope time difference EITD may be further input to the artificial neural network unit 13.

In the change of spatial feeling, the degradation caused by the change of the position of the sound image is one of the important evaluation factors. According to the classical Duplex theory, the position of the sound image can be recognized by the magnitude difference ( ILD ) for the high frequency component. However, recent studies have shown that not only the magnitude difference ( ILD ) but also the time difference ( EITD ) of the high-frequency envelope influences the recognition of sound images.

In an embodiment of the present invention proposes a method for the amount of how the amount of the high-frequency component calculate the size difference (ILD) and an envelope calculating a time difference (EITD).

In order to evaluate objective performance of multi-channel audio, quantitative analysis of tone distortion and quantitative analysis of spatial distortion are required. One of the important factors in evaluating the spatial distortion is the distortion of sound image orientation. Humans are quantitative on the basis of these two characteristics because the amount of interfering difference ( ILD ) and the amount of envelope use temporal difference ( EITD ) in recognizing the sound image of high frequency components. We can evaluate the sound localization performance. The positive magnitude difference ( ILD ) and the positive time difference ( EITD ) of the envelope are used for reference signals (e.g., the original sound) and test signals (e.g., the sound encoded and decoded by the codec). The ILD distortion ( ILDDist ) and the EITD distortion ( EITDDist ) values may be calculated as the cognitive distances of the difference between the reference signal and the test signal. In order to calculate ILD and EITD of high frequency components, multi-channel sound sources must first be synthesized with positive signals. Head-Related Transfer Functions (HRTFs) are used for bipolar signal synthesis, where HRTF refers to the sound propagation path from each speaker position to both ears. The amount synthesized in this way can be used to calculate the high frequency component ILD and EITD .

6 is a flowchart for calculating the above-described ILD distortion.

)와 테스트 신호의 양이 입력 신호(

)를 산출할 수 있다. 도 6의 말초 청각 모형부(pheripheral ear model)(602)에서는 기준 신호의 양이 입력 신호(

)와 테스트 신호의 양이 입력 신호(

)를 각각 입력받아 기준 신호의 자극 패턴(excitation pattern)과 테스트 신호의 자극 패턴을 각각 산출할 수 있다. 도 6의 포락선 추출부(envelop extraction)(603)에서는 기준 신호의 자극 패턴과 테스트 신호의 자극 패턴을 입력받아 기준 신호의 자극 패턴의 포락선과 테스트 신호의 자극 패턴의 포락선을 각각 산출할 수 있다. 도 6의 인지 모델부(cognition model)(604)에서는 기준 신호의 자극 패턴의 포락선과 테스트 신호의 자극 패턴을 입력받아 고주파 성분의 ILDDist를 산출할 수 있다.In the Binaural Synthesis 601 of FIG. 6, the aforementioned LF _test , RF _test , C _test , LS _test , RS _{test ,} and LF _ref , RF _ref , C _ref , LS _ref , and RS _ref are respectively input. Take the reference signal amount as the input signal (

) And the amount of test signal

) Can be calculated. In the peripheral ear model 602 of FIG. 6, the amount of reference signal is determined by the input signal (

) And the amount of test signal

) Can be respectively input to calculate an excitation pattern of the reference signal and an excitation pattern of the test signal. The envelope extraction unit 603 of FIG. 6 may receive the stimulus pattern of the reference signal and the stimulus pattern of the test signal to calculate an envelope of the stimulus pattern of the reference signal and an envelope of the stimulus pattern of the test signal, respectively. In the recognition model 604 of FIG. 6, an ILDDist of a high frequency component may be calculated by receiving an envelope of a stimulus pattern of a reference signal and a stimulus pattern of a test signal.

The amount synthesizing unit 601 of FIG. 6 may correspond to the preprocessor 11 of FIG. 2. The peripheral auditory model 602, the envelope extractor 603, and the cognitive model 604 of FIG. 6 may be included in the output variable calculator 12 of FIG. 2.

ILD can be defined as the energy difference of the ear input signal passing through the peripheral ear model, which consists of a bandpass filter with a center frequency on the ERB scale, which can be expressed as have. The peripheral auditory model is an auditory model that calculates an excitation pattern occurring in the basallar membrane from signals input to both ears.

&Quot; (3) "

Although the energy difference between the input signals of both ears can be expressed as Equation 3, the mechanism generated in the brain by the actual ILD can be different. If the energy of the input signal is different, an Inferior Colliculus (IC) that processes the ILD may need to deal with the neural spike in the ear having a large input. Since the model for the number of neural spikes occurring in the IC follows the form of a tangent sigmoid function, the calculated ILD values are nonlinearly transformed by the colon function, which is represented by equation (4) and (4). It can be expressed as 5.

&Quot; (4) "

[Equation 5]

In this case, the slope (S) of the colon function has a different sign according to the energy difference of the ear input signal, which has a positive value when the input of the left ear is large and a negative value when the input of the right ear is large. Can be. In addition, to reflect the sensitivity of the neural spike generation mechanism in the IC according to each frequency band may have a different slope for each band. Tk is the threshold of the colon function, which is 0 for ILD . Thereafter, the ILD distortion value for the time-frequency divided signal may be calculated as shown in Equation 6.

&Quot; (6) "

The final ILD distortion can be calculated by obtaining an average value over a frequency band and a time frame, which can be expressed as Equation (7). The final ILD distortion can be regarded as the cognitive distance between the test signal and the reference signal due to the ILD .

[Equation 7]

The EITD distortion represents the cognitive distance between the difference in the image position between the test source and the reference source caused by the difference in the envelope difference time difference. EITD distortion can be used as an element to evaluate the spatial feeling caused by the difference in the image position of high frequency sound source with ILD distortion.

7 is a diagram illustrating a flow of calculating EITD distortion.

)와 테스트 신호의 양이 입력 신호(

)를 산출할 수 있다. 양이 합성부(701)는 도 2의 전처리부(11)에 대응될 수 있다. In the Binaural Synthesis 701 of FIG. 7, the above-described LF _test , RF _test , C _test , LS _test , RS _{test ,} and LF _ref , RF _ref , C _ref , LS _ref , and RS _ref are respectively input. Take the reference signal amount as the input signal (

) And the amount of test signal

) Can be calculated. The amount synthesizer 701 may correspond to the preprocessor 11 of FIG. 2.

,

로 표시할 수 있다. 멀티채널 음원 및 양이 신호에서 아래첨자 test와 ref는 각각 평가 신호와 참조신호를 의미한다.The positive synthesizer 701 synthesizes a multichannel sound source using positive head signals using head related transfer functions (HRTFs), respectively.

,

As shown in FIG. The subscripts test and ref in the multichannel sound source and positive signal mean the evaluation signal and the reference signal, respectively.

The head transfer function used for quantitative signal synthesis is ITU-R Rec. BS. It may be recorded in a standard environment auditorium as recommended in 1116-1, and the LFE channel may be tuned to zero for all sources. Equation 8 may be used to synthesize a bi-signal from five channel signals.

[Equation 8]

,

은 각각 양쪽 귀의 입력 신호를 의미한다.In Equation _{8, H CL, H LfL,} H RfL, H LsL, H RsL, H CR, H LfR, H RfR, H LsR, H RsR is ten quantity representing a sound wave transmission path to both ears from each of the speaker Are the Binaural Room Transfer Functions (BRTFs),

,

Are the input signals of both ears, respectively.

The positive signal synthesized in this manner can be processed by a peripheral ear model. In fact, the input signal coming into both ears is processed by the cochlea via the middle ear, which simulates this process. Cochlea simulators in peripheral auditory models convert bilateral input signals into signals that stimulate hair cells in the human basal membrane. The cochlear simulator can be thought of as a filter bank consisting of 24 bandpass filters with a center frequency determined by the Equivalent Rectangular Bandwidth (ERB) scale, with the signal passing through the simulator being the stimulus of the signal passing through each bandpass filter. Can be converted into an excitation pattern.

)와 테스트 신호의 양이 입력 신호(

)를 각각 입력받아 기준 신호의 자극 패턴(excitation pattern)과 테스트 신호의 자극 패턴을 각각 산출할 수 있다. In the peripheral ear model 702 of FIG. 7, the amount of reference signal is determined by the input signal (

) And the amount of test signal

) Can be respectively input to calculate an excitation pattern of the reference signal and an excitation pattern of the test signal.

The envelope extraction unit 703 of FIG. 7 receives the stimulus pattern of the reference signal and the stimulus pattern of the test signal to calculate the envelope of the stimulus pattern of the reference signal and the stimulus pattern of the test signal, respectively.

An envelope of the stimulus pattern may be extracted by applying a discrete Hilbert transform to components of the high frequency region of the transformed stimulus pattern. FIG. 8 illustrates an example of envelope extraction. The solid line represents the full-rectified excitation pattern, and the dotted line represents the extracted envelope. The amount of envelope extracted can calculate the time difference to obtain the EITD .

The signal coming into the output after the positive signal passes through the ERB-scale auditory filter bank may be a signal (x [k, n] or X [k, n]) segmented in the time-frequency domain. Where k is a frequency band number and n is a time frame number. Using the discrete Hilbert transformed value H {x [k, n]} from this signal, the envelope E [k, n] of the signal can be calculated as shown in Equation (9). x [k, n] may be represented by r [k, n], and H {x [k, n]} may be represented by i (n).

[Equation 9]

In Equation 9, k denotes a frequency band index divided by a peripheral ear model, and n denotes a time frame index to be processed.

In the recognition model unit 704 of FIG. 7, the envelope of the stimulus pattern of the reference signal and the envelope of the stimulus pattern of the test signal may be input to calculate the EITD distortion of the high frequency component.

The extracted ITD ( EITD ) of the envelope can be calculated using a normalized cross-correlation function (NCF) for input signals of both time-frequency-divided ears, which can be expressed as have.

[Equation 10]

In Equation 10, E _{L , k, n} , E _{R , k, n} are envelope signals of stimulus patterns occurring at both ears, d is time delay in sample units, k is frequency band index, and n means time frame index. E _{L , k, n} , E _{R , k, n} may be represented by X _{L , k, n} , X _{R , k, n} , respectively.

Envelope InterAural Cross-correlation Coefficient ( EIACC ) is defined as the maximum value of NCF, and Envelope Interaural Time Difference ( EITD ) is the time delay value when NCF has the maximum value. It can be defined as. EITD and EIACC may be calculated by the NCF, and may be calculated as shown in Equations 11 and 12 for the time-frequency divided signals, respectively.

[Equation 11]

[Equation 12]

In Equations 11 and 12, the parameter N is in the range of d , and the theoretically possible amount means the value of the time difference. EITD and EIACC are calculated for the reference and evaluation signals, respectively, and are also indicated by the subscripts ref and test . In addition, since the cognitive difference of the sound source direction based on the high frequency envelope may be approximated by the distance between two points on the unit circle, the difference between the EITD of the reference signal and the test signal may be calculated as shown in Equation 13. That is, the EITD difference occurred between the test signal and the reference signal may be calculated as the difference between the two vectors on the unit circle having a phase angle corresponding to EITD. In Equation 13, f _s represents a sampling frequency.

[Equation 13]

After calculating Δ EITD in this way, it should be taken into account that the cognitively unsuccessful phase misalignment occurs. If the EIACC is very low, the direction of the sound source perceived by the EITD is not clear, so a decision factor can be applied to the difference value of the EITD to take into account the confidence in detecting the perceived sound source direction. have. To model confidence, we can use the tangent sigmoid function, which transforms the EIACC values nonlinearly. That is, EIACC value can be non-linear conversion using an S-shaped colon EIACC value tangent function to account for may not be aware of the location of the sound source is too low. For the reference signal and the evaluation signal, the EIACC value may be nonlinearly converted by Equations 14 and 15.

[Equation 14]

[Equation 15]

In equations (14) and (15), s and T _k represent the slope and threshold of the colon function, respectively. For EITD , the slope is 50 and the threshold can use different values for each band to reflect sensitivity to the EITD in each band. have.

The EITD distortion after the determinant is applied to Δ EITD may be calculated as shown in Equation 16. That is, the EITD distortion may be calculated using the nonlinear transformed EIACC value as a determinant.

[Equation 16]

The final EITD distortion can be obtained by taking an average value over the entire frequency band and time frame as shown in Equation 17. In other words, distortion EITD had an average EITD distortion, which means that the distances of the sound source position relative to the reference signal and the test signal due to the difference EITD.

[Equation 17]

The above-described peripheral auditory model 702, envelope extractor 703, and cognitive model 704 of FIG. 7 may be included in the output variable calculator 12 of FIG. 2.

9 illustrates the flow of calculating EITD distortion in accordance with FIG. 7 in more detail.

The positive synthesizing unit 901 may synthesize a multichannel sound source as a positive signal in the same manner as in Equation (8). The peripheral auditory model unit 902 may receive the input signal from the amounts of the reference signal and the test signal to calculate the stimulation pattern of the reference signal and the stimulation pattern of the test signal. The envelope extractor 903 may calculate an envelope of each signal as shown in Equation (9). The cross correlation function applying unit 904 may calculate the EITD and the EIACC using an envelope. In EITD distortion calculation section 905, using the EITD EIACC and the test signal and the reference signal it is possible to calculate the distortion value EITD. In FIG. 9, the subscripts R, L, test, ref, k, and n represent a right channel, a left channel, a test signal, a reference signal, a frequency band index, and a time frame index, respectively.

The method for obtaining the EITD distortion by Equations 13 to 17 may be modified in a manner using Equations 18 to 19 below.

Before calculating the EITD distortion value corresponding to the cognitive distance of the EITD from the EITD values calculated for the reference signal and the test signal, respectively, as shown in Equation 11, the tangent S letter to the EIACC value as shown in Equations 14 and 15 The colon function can be used to nonlinearly convert EIACC values.

In this way, the nonlinear transformed EIACC value may be used as a weight value of the EITD . The distance of the cognitive EITD can be calculated from the weighted EITD value. In this case, the difference in the sound source direction due to the weighted EITD may be expressed as an Euclidian distance between two points on the unit circle, which may be calculated as shown in Equation 18. In Equation 18, c _test [k, n] and c _ref [k, n] may be represented by p _test [k, n] and p _ref [k, n], respectively.

Equation 18

The final EITD distortion for the entire sound source can be calculated by taking the average over the entire frequency band and time frame as shown in Equation 19. The final EITD distortion may represent the average EITD distortion, which means the cognitive distance between the reference signal and the test signal due to the EITD difference.

[Equation 19]

The sound quality evaluation apparatus according to an embodiment of the present invention is based on a multi-channel audio signal input from each channel (L, R, C, LS, RS) of the multi-channel audio reproduction system (Binaural input signal) Preprocessing means for generating a signal, the generated amount of the input signal includes a degree of correlation distortion ( ICACCist : IACC Distortion), a magnitude difference distortion ( ILDDist : ILD Distortion), the envelope difference time difference distortion ( EITDDist ) Output variable calculation means for calculating the model output variable, and the artificial neural network means for outputting the grade of sound quality based on the model output variable.

In order to examine the effect of spatial factors on objective sound quality evaluation for multi-channel sound sources, subjective listening evaluation was conducted first. The subjective audit assessment database used in this experiment was distributed by the ISO / MPEG Audio Group and is available in ITU-R Rec. BS. Implemented as recommended in 1534-1 "Multiple Stimulus with Hidden Reference and Anchor" (MUSHRA). Eleven kinds of sound sources were used for subjective listening evaluation. Each sound source was encoded and decoded using eleven different multichannel audio coding techniques, resulting in 121 items through subjective audit evaluation.

Table 1 shows the correlation coefficient between the subjective listening evaluation results and the 14 evaluation factors used for the objective evaluation.

Correlation coefficient between subjective listening evaluation result and 14 factors Evaluation factor Correlation coefficient ADB -0.68 NMRtoB -0.51 NLoundB -0.51 AModDif1B -0.45 WModDif1B -0.44 RDF -0.43 EHS -0.43 AModDif2B -0.36 AvgBwRef -0.06 AvgBwTst -0.00 ILDD -0.78 IACCD -0.62 ITDD -0.61 EITDD -0.72

Each correlation coefficient ρ _{X, Y} is calculated as shown in Equation 20.

[Equation 20]

In Equation 20, X denotes MOS and Y denotes data of each factor, and the correlation coefficient between 14 factors and subjective listening evaluation was calculated for 121 signals synthesized with positive signals. The first 10 of the 14 evaluators are now ITU-R Rec. BS. This is the model output variable used in 1387-1. Ten model output variables and four spatial factors were summarized in Tables 2 and 3, respectively.

<ITU-R Rec. BS. Model Output Variables Used as Factors for Tone Degradation in 1387-1 factor Explanation ADB Averaged distortion block. Ratio of total distortion to total number of distorted blocks NMRtotB Logarithmic value of masker energy to average noise ratio EHS Overtone structure of error BWRef The bandwidth of the reference signal Bwtest Evaluation Signal Bandwidth AModDif1B Average modulation difference AModDif2B Average Modulation Difference Weighted Modulation Variations When Reference Signals Have Little Modulation WinModDifB Average Modulation Difference with Window Function RDF Number of frames with noise-based energy ratio above the reference NLoudB Average noise strength

<Both elements indicating deterioration of space> factor Explanation ITDDist Cognitive distance of sound source direction difference between evaluation signal and reference signal caused by positive time difference ILDDist Cognitive distance of sound source direction difference between evaluation signal and reference signal caused by difference in magnitude IACCDist Cognitive distance of the difference in sound source wideness between the evaluation signal and the reference signal caused by positive correlation coefficient aberration EITDDist Cognitive distance of sound source direction difference caused by time difference between high frequency region envelope

Since all factors have a negative correlation with the subjective listening evaluation, the greater the absolute value of the correlation coefficient in Table 1, the better the sound quality prediction performance. As can be seen from Table 1, EITDDist has a very high correlation with the subjective listening evaluation result with a correlation coefficient of 0.72. In particular, EITDDist may have a higher correlation than ITDDist having a subjective listening evaluation results and the correlation coefficient of 0.62 the IACCDist or the correlation coefficient of 0.61 which can be confirmed by having a higher than the correlation existing 10 voice distortion factor. From these results, it can be seen that the high frequency envelope information plays an important role in the perception of spatiality and overall sound quality by the multi-channel audio signal. In addition, existing ITU-R Rec. BS. Compared with the tone component used in 1387-1, the subjective audit evaluation result and the four spatial components can be found to have a similar or higher correlation coefficient. Based on these results, it can be seen that not only the timbre but also the spatial elements are important in the quality evaluation of multichannel audio.

Each factor calculated in this way may be used as an input factor of a prediction model for objective sound quality evaluation of multichannel audio. It is possible to obtain better prediction performance when the spatial distortion factor with high correlation coefficient is added as an input factor to the objective sound quality prediction model of the multichannel audio coding system. EITDDist can be used as a factor for evaluating spatial distortion in an objective sound quality prediction model. In particular, since the EITDDist has a high correlation with the subjective listening evaluation results, the performance of the objective sound quality prediction model of the multi-channel audio coding system can be improved by adding the EITDDist as an input factor of the sound quality prediction model.

Through the present invention, it is possible to increase the performance of the evaluation by providing a sense of space for the objective evaluation of the multi-channel sound source. Each spatial factor can be used to mathematically model the processes in which the auditory signal is processed in the real brain to generate an evaluation model that can reflect cognitive differences.

In embodiments of the present invention, the artificial neural network unit 13 may be replaced with a general digital signal processing unit. That is, the artificial neural network unit 13 is presented as an example of the digital signal filter in order to explain the embodiment of the present invention. Accordingly, the scope of the invention is not limited by the accompanying drawings and the description thereof.

According to the present invention, based on psychoacoustic and physiological research results, finding factors that influence spatial cognition and implementing the corresponding factors by mathematical models can improve the performance of the objective evaluation model of the multi-channel audio codec.

The method of the present invention as described above may be embodied as a program and stored in a computer-readable recording medium (such as a CD-ROM, a RAM, a ROM, a floppy disk, a hard disk, or a magneto-optical disk). Since this process can be easily implemented by those skilled in the art will not be described in more detail.

The embodiments described above are the components and features of the present invention are combined in a predetermined form. Each component or feature is to be considered optional unless stated otherwise. Each component or feature may be implemented in a form that is not combined with other components or features. It is also possible to combine some of the components and / or features to form an embodiment of the invention. The order of the operations described in the embodiments of the present invention may be changed. Some configurations or features of certain embodiments may be included in other embodiments, or may be replaced with corresponding configurations or features of other embodiments. It is obvious that the claims may be combined to form an embodiment by combining claims that do not have an explicit citation relationship in the claims or as new claims by post-application correction.

Embodiments according to the present invention may be implemented by various means, for example, hardware, firmware, software, or a combination thereof. In the case of a hardware implementation, an embodiment of the present invention may include one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), FPGAs ( Field Programmable Gate Arrays), processors, controllers, microcontrollers, microprocessors, and the like.

In the case of implementation by firmware or software, an embodiment of the present invention may be implemented in the form of a module, procedure, function, etc. that performs the functions or operations described above. The software code may be stored in a memory unit and driven by a processor. The memory unit may be located inside or outside the processor, and may exchange data with the processor by various known means.

It will be apparent to those skilled in the art that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. Accordingly, the above detailed description should not be interpreted as limiting in all aspects and should be considered as illustrative. The scope of the invention should be determined by reasonable interpretation of the appended claims, and all changes within the equivalent scope of the invention are included in the scope of the invention.

The present invention described above is capable of various substitutions, modifications, and changes without departing from the technical spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited by the drawings.

The present invention is different from simply quantifying the degree of distortion of the energy of the recovered signal after compression, compared to the original signal, and statistically processing the auditory evaluation felt by the selected listener in a multi-channel audio reproduction environment. To achieve similar results. This eliminates the listening evaluation and statistical processing of the sound quality of the multi-channel audio compression codec, and can perform the sound quality evaluation only through the measurement or predict the sound quality evaluation result.

An embodiment of the present invention provides a method for evaluating the performance of an audio compression codec by objectively comparing the perceived quality of a reproduced sound formed by encoding and then decoding a reference signal by using an audio compression codec. Can be used for devices.

Claims (13)

One or more model output variables (MOVs) including variables representing an envelope interaural time difference ( EITDDist , Envelope Interaural Time Difference Distortion) by comparing a reference signal and a signal under test. Producing); And
Mapping the at least one model output variable to a value corresponding to audio quality;
Including,
How to measure sound quality. The method of claim 1,
EITDDist, which is a variable representing the envelope lag distortion,

Given by

Is an envelope positive time difference distortion generated by comparing the reference signal with a k-th frequency band of an n-th time frame of the test signal. The method of claim 2,
remind

Is

Given by
remind

Denotes a difference between an envelope difference time difference ( EITD ) in a k-th frequency band of an n-th time frame of the reference signal and the test signal,
remind

Is the nonlinear transform value of the envelope coefficient in the k th frequency band of the n th time frame of the test signal ( EIACC , Envelope InteAural Cross-correlation Coefficient),
remind

Is the nonlinear transform value of the envelope coefficient in the k th frequency band of the n th time frame of the reference signal ( EIACC , Envelope InteAural Cross-correlation Coefficient),
How to measure sound quality. The method of claim 1,
And the reference signal is generated from a multichannel audio signal, and the test signal is generated by passing the multichannel audio signal to a device under test for measuring the sound quality. The method of claim 1,
At least one of the one or more model output variables is generated by comparing an excitation pattern of the reference signal and the test signal. The method of claim 1,
The variable representing the envelope lag distortion is generated by passing the reference signal and the test signal through a filter bank. On your computer,
Comparing the reference signal with the test signal to generate one or more model output variables including a variable representing an envelope lag distortion; And
Mapping the at least one model output variable to a value corresponding to audio quality;
A computer-readable medium having recorded thereon a program for executing the program. The method of claim 7, wherein
EITDDist, which is a variable representing the envelope lag distortion,

Given by

Is an envelope positive time difference distortion generated by comparing the reference signal with a k-th frequency band of an n-th time frame of the test signal. A computer-readable medium that records a code for changing a program that maps one or more model output variables generated by comparing a reference signal with a test signal to values corresponding to sound quality.
The code is adapted to modify the program such that a variable representing an envelope positive time difference distortion obtained by comparing the reference signal and the test signal is included in the one or more model output variables.
Computer-readable media. 10. The method of claim 9,
EITDDist, which is a variable representing the envelope lag distortion,

Given by

Is an envelope positive time difference distortion generated by comparing the reference signal with a k-th frequency band of an n-th time frame of the test signal. One or more model output variables (MOVs) including variables representing an envelope interaural time difference ( EITDDist , Envelope Interaural Time Difference Distortion) by comparing a reference signal and a signal under test. Model output variable generating means for producing (); And
Mapping means for mapping the at least one model output variable to a value corresponding to audio quality;
Including,
Sound quality measuring device. The method of claim 11,
Wherein said generating means and said mapping means are part of a processing device adapted to drive a program for executing said generating and said mapping step. The method of claim 11,
EITDDist, which is a variable representing the envelope lag distortion,

Given by

Is an envelope positive time difference distortion generated by comparing the reference signal with a k-th frequency band of an n-th time frame of the test signal.

Images