Sound quality evaluation method, device and computer storage medium

Abstract

The present invention discloses a sound quality evaluation method, device and computer storage medium. The sound quality evaluation method comprises: obtaining music to be evaluated after the parameters of the equalizer to be tested are adjusted; inputting the music to be evaluated into a pre-trained neural network; obtaining the evaluation result of the sound quality of the music to be evaluated under the parameters of the equalizer to be tested output by the neural network. It can be seen that the embodiment of the present invention uses a pre-trained neural network to evaluate the sound quality of music, which is simple to operate, does not require the organization of an expert team, and the results are more accurate and reliable.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of sound, and more specifically, to a sound quality evaluation method, device and computer storage medium. Background Art

[0002] The quality of sound depends on many factors. In order to evaluate the quality of reproduced sound, a sound quality evaluation method can be used. Generally, the sound quality evaluation method includes two methods: subjective evaluation and objective evaluation.

[0003] The subjective evaluation method is to organize an expert team to score the sound quality, but this method is too expensive. The objective evaluation method is to use instruments to test the technical indicators of the sound, but this measurement method is quite different from the subjective listening method, which will result in the sound quality of the sound with a perfect frequency response curve being not good.

[0004] Therefore, it is currently difficult to evaluate the sound quality of reproduced sound. Summary of the invention

[0005] The present invention provides a sound quality evaluation method, device and computer storage medium, which can facilitate sound quality evaluation and produce accurate and reliable results.

[0006] According to a first aspect of the present invention, there is provided a sound quality evaluation method, comprising:

[0007] Obtain the music to be evaluated after the parameters of the equalizer to be tested are adjusted;

[0008] Inputting the music to be evaluated into a pre-trained neural network;

[0009] Obtain an evaluation result of the neural network output on the sound quality of the music to be evaluated under the equalizer parameters to be tested.

[0010] In one implementation of the present invention, the music to be evaluated is input into a pre-trained neural network, including: performing Fourier transform on the music to be evaluated to obtain a feature vector of the music to be evaluated; and inputting the feature vector of the music to be evaluated into the pre-trained neural network.

[0011] In one implementation of the present invention, the neural network is trained in the following manner: constructing a training set, the training set including multiple training data and sound quality evaluations corresponding to the multiple training data, each training data including audio information and corresponding equalizer parameters, and the corresponding sound quality evaluation is an evaluation of the sound quality of the audio information under different equalizer parameters; training the neural network based on the training set.

[0012] In one implementation of the present invention, the construction of the training set includes: performing a plurality of different equalizer parameter adjustments on a certain audio information, and obtaining a certain person's sound quality evaluation of the audio information after the plurality of different equalizer parameter adjustments.

[0013] In one implementation of the present invention, the training of the neural network based on the training set includes: preprocessing the training data in the training set to obtain preprocessed data; extracting feature vectors of the preprocessed data; and inputting the feature vectors into the neural network for training.

[0014] In an implementation of the present invention, the preprocessing includes: truncating the training data in the training set according to a predetermined time length to obtain data with multiple predetermined time lengths.

[0015] In an implementation of the present invention, extracting the feature vector of the preprocessed data includes: performing fast Fourier transform, overlapping, and Mel spectrum transform on the preprocessed data to obtain the feature vector of each data.

[0016] In a second aspect, a sound quality evaluation device is provided, the device being used to implement the steps of the method described in the first aspect or any implementation manner, the device comprising:

[0017] An acquisition module, used for acquiring the music to be evaluated after the parameters of the equalizer to be tested are adjusted;

[0018] An input module, used for inputting the music to be evaluated into a pre-trained neural network;

[0019] An output module is used to obtain the evaluation result of the neural network output on the sound quality of the music to be evaluated under the equalizer parameters to be tested.

[0020] In a third aspect, a sound quality evaluation device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of the method described in the first aspect or any implementation manner are implemented.

[0021] In a fourth aspect, a computer storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in the first aspect or any implementation manner are implemented.

[0022] It can be seen that in the process of constructing the training set, the embodiment of the present invention takes into account both objective and subjective methods, and uses deep learning technology for training, thereby obtaining a reliable neural network for evaluating sound quality. The sound quality evaluation obtained by the neural network is not only low-cost and does not require the organization of a large number of expert teams, but also the results are accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above and other purposes, features and advantages of the present invention will become more apparent by describing the embodiments of the present invention in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0024] Figure 1 It is a scene diagram of the frequency response curve of headphones tested by prior art instruments;

[0025] Figure 2 It is a scene diagram in which a team of existing technical experts scores the sound quality of headphones;

[0026] Figure 3 is a schematic diagram of training a neural network according to an embodiment of the present invention;

[0027] Figure 4 is a schematic diagram of the training accuracy of an embodiment of the present invention;

[0028] Figure 5 is a schematic diagram of the training error of an embodiment of the present invention;

[0029] Figure 6 is a schematic diagram of the verification accuracy of an embodiment of the present invention;

[0030] Figure 7 is a schematic flow chart of a sound quality evaluation method according to an embodiment of the present invention;

[0031] Figure 8 is a schematic block diagram of a sound quality evaluation device according to an embodiment of the present invention;

[0032] Fig. 9 is another schematic block diagram of the sound quality evaluation device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical scheme and advantages of the present invention more obvious, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described in the present invention, all other embodiments obtained by those skilled in the art without creative work should fall within the protection scope of the present invention.

[0034] The quality of sound (such as music, etc.) depends on many factors, including but not limited to: sound source characteristics, signal characteristics of audio equipment, sound field characteristics, auditory characteristics, etc. Among them, sound source characteristics may include, for example, sound pressure, frequency, spectrum, etc.; signal characteristics of audio equipment may include, for example, distortion, frequency response, dynamic range, signal-to-noise ratio, transient characteristics, stereo separation, etc.; sound field characteristics may include, for example, direct sound, early reflected sound, reverberation sound, interaural correlation coefficient, reference vibration, sound absorption rate, etc.; auditory characteristics may include, for example, loudness curve, audible range, various hearing sensations, etc. Therefore, it is difficult to evaluate sound quality.

[0035] The objective evaluation method is to use instruments to test technical indicators, for example, the frequency response curve of equipment (such as headphones) can be tested. For example, a scene diagram of testing the frequency response curve of headphones is as follows: Figure 1 shown. Figure 1 In the test, the instrument used for testing includes a test part, test suite 1, test suite 2 and test suite 3, wherein test suite 3 is a signal generator, test suite 2 is the main part of the test, and test suite 1 is a receiver. Specifically, a sweep frequency signal, such as 200Hz to 8kHz, can be generated by a computer and input to the headphones to be tested via a signal generator, and the output of the headphones is tested to obtain a frequency response curve. Furthermore, the obtained frequency response curve can be compared with the target frequency response curve, so as to objectively judge the quality of the sound.

[0036] The subjective evaluation method is to organize an expert team and have the expert team give scores. For example, a scene diagram such as Figure 2 shown. Figure 2 In the process, the computer generates sound, and then the equalizer (eq) parameters of the sound can be adjusted differently to obtain several adjusted sounds, which are output through headphones and scored by a team of experts to determine which eq parameters correspond to the highest sound quality, such as Figure 2 The one located above. Figure 2 In the game, the computer generates sound and outputs it through different headphones. The expert team scores to determine which headphones have the best sound quality. Figure 2 The one located below.

[0037] However, the two evaluation methods in the prior art each have their own defects, such as being greatly different from actual conditions and being too costly.

[0038] The embodiment of the present invention provides a neural network, which can evaluate the sound quality of music and determine the quality of the music under different EQ parameters.

[0039] The embodiment of the present invention can construct a training set, and train a neural network based on the training set, thereby obtaining a trained neural network, which can be used for subsequent evaluation of the sound quality of music.

[0040] The training set includes a plurality of training data and corresponding sound quality evaluations, wherein the training data may include audio information and corresponding equalizer parameters, and the sound quality evaluation is an evaluation of the sound quality of the audio information under different equalizer parameters.

[0041] For example, the following table 1 is an example of training data and corresponding sound quality evaluation. However, it is understandable that table 1 is only for illustration, and the present application does not limit the specific form of training data.

[0042] Table 1

[0043]

[0044] Among them, the sound quality evaluation corresponding to the training data can be obtained through subjective evaluation, so the accuracy is high, thereby ensuring the availability of the neural network. Referring to Table 1, the sound quality evaluation of the same audio information (such as audio information 1) under different equalizer parameters (such as eq parameter 1, eq parameter 2, eq parameter 3) may be different, and the sound quality evaluation of different audio information (such as audio information 1, audio information 2) under the same equalizer parameters (such as eq parameter 3) may be different.

[0045] In addition, it is understandable that the sound quality evaluation can be used to indicate the degree of preference, such as including like, medium, and dislike in Table 1. However, the present application is not limited to this, for example, more or fewer evaluation levels can be divided, and for example, evaluation can be performed in other ways (such as scoring), and so on.

[0046] The following first describes how to obtain the sound quality evaluation corresponding to a certain training data, and then construct a training set.

[0047] For a certain audio information (for the convenience of description, it is assumed to be music M below), the corresponding sound quality evaluation under different equalizer parameters can be obtained in the following way.

[0048] (1) For music M, a variety of eq parameters can be adjusted, and then played through the same equipment to obtain a person's sound quality evaluation of the music M after the various eq parameters are adjusted. The sound quality evaluation can be a person's preference for the eq parameters.

[0049] Exemplarily, 8 different eq parameter adjustments can be made to obtain 8 corresponding music of the corresponding music M, and then a certain person can like the 8 corresponding music. The degree of preference can include like, medium, and dislike. Or, the degree of preference can include very like, generally like, medium, and dislike. And so on, that is, the degree of preference can be 3 categories, 4 categories, or even 8 categories. For example, the 8 corresponding music can be sorted by the degree of preference, that is, it can be understood that the degree of preference is divided into 8 categories.

[0050] Optionally, the embodiment of the present invention assumes that the preference level includes like, medium, and dislike. Then, when determining the preference level, a blind evaluation of 4, 2, and 1 elimination round can be used for scoring.

[0051] Specifically, the 8 corresponding music after the music M is adjusted with 8 different eq parameters can be numbered, which can be No1, No2, No3, No4, No5, No6, No7, and No8 in sequence. Then, it can be divided into 4 groups, each group includes 2 corresponding music, and let a certain person listen to it, so as to obtain 4 disliked ones. For example, No1 and No2 are a group, and the better one can be selected from them, and the other eliminated is "disliked"; No3 and No4 are a group, and the better one can be selected from them, and the other eliminated is "disliked"; No5 and No6 are a group, and the better one can be selected from them, and the other eliminated is "disliked"; No7 and No8 are a group, and the better one can be selected from them, and the other eliminated is "disliked". In this way, 4 non-eliminated ones are selected and divided into 2 groups, and each group selects a better one and eliminates the other. Among them, the one eliminated in each group is "medium". After that, the selected two can be compared again, and the better one can be selected, so the one eliminated this time is "medium" and the selected one is "liked". The above process was carried out three times, the first time it was divided into 4 groups, the second time it was divided into 2 groups, and the third time it was divided into 1 group, which can be called the "4, 2, 1" elimination scoring system. In this way, for the 8 corresponding music after 8 different eq parameter adjustments, 4 were disliked, 3 were average, and 1 was liked.

[0052] (2) For music M, it can be played through a variety of equipment to obtain a person's evaluation of the sound quality of the various equipment. The sound quality evaluation can be a person's preference for equipment.

[0053] Among them, the frequency response curve of each equipment can be obtained through testing, so it can be understood that a person’s preference for various equipment is equivalent to obtaining a person’s preference for frequency response curves.

[0054] By matching the results obtained in (1) and (2) above, we can obtain the corresponding relationship between a person's favorite frequency response curve and the eq parameter.

[0055] Furthermore, based on the eq parameter preferred by a certain person obtained in (1), the music samples of the same type (or style) can be processed to obtain the music preferred by the person after the eq parameter adjustment. For example, assuming that the style of the music M is folk music, and the eq parameter preferred by a certain person in (1) is eq parameter 1, then the eq parameter 1 can be adjusted for other (assuming M1, such as 100 or more or less) sample music of the same folk music.

[0056] For another example, assuming that the style of the music M is folk music, and the eq parameters that a certain person dislikes in (1) are eq parameter 2, eq parameter 3, eq parameter 4, and eq parameter 5, then the eq parameter 2, eq parameter 3, eq parameter 4, and eq parameter 5 can be adjusted for other samples of folk music (assuming M1, such as 100 or more or less) to obtain 400 music samples that are disliked by the certain person. If the eq that a certain person likes in (1) is medium is eq parameter 6, eq parameter 7, and eq parameter 8, then the eq parameter 6, eq parameter 7, and eq parameter 8 can be adjusted for other samples of folk music (assuming M1, such as 100 or more or less) to obtain 300 music samples that are medium to the certain person.

[0057] In this way, the embodiment of the present invention obtains the first training data corresponding to the plurality of audio information.

[0058] Assuming that the above processes (1) and (2) are performed on a certain person (assuming it is P1), then through the above process, the sound quality evaluation of a certain person (assuming it is P1) for a certain type of audio information under different eq parameters can be obtained, that is, multiple training data and corresponding sound quality evaluations can be obtained.

[0059] Similarly, the above process can be performed for other people to obtain more training data and corresponding sound quality evaluations. For example, the same process can be performed for P2, P3, P4, etc. to obtain the sound quality evaluations of these people on a certain type of audio information at various eq parameters.

[0060] For the convenience of description, it is assumed that there are 100 people here, so that for music M, the preferences of 100 people can be obtained respectively. By combining the preferences of 100 people, the preferences of most people can be obtained. For example, if the eq parameter with the largest number of "likes" (such as 50 people, 80 people, etc.) among 100 people is eq parameter 2, then for music M, eq parameter 2 is the preference of most people.

[0061] Furthermore, the same type of music samples can be processed according to the obtained eq parameters liked by most people (such as eq parameter 2 in the above example) to obtain the music liked by the individual after the eq parameters are adjusted. For example, assuming that the style of the above music M is folk music, and the eq parameter liked by most people is eq parameter 2, then the eq parameter 2 can be adjusted for other several (assuming M1, such as 100 or more or less) sample music of the same folk music.

[0062] In this way, the embodiment of the present invention obtains the second training data corresponding to the plurality of (M1) audio information.

[0063] Through the process described above, for music M, the corresponding sound quality evaluation can be obtained by having multiple people (such as 100 people) subjectively mark their preferences, and combined with the frequency response curve of the equipment obtained through objective testing, the first training data and the second training data can be obtained. Similarly, for other music (such as another type of music), the corresponding first training data and the second training data can also be obtained through the above process.

[0064] That is to say, assuming that there are multiple different types of audio information, you can select a piece of music from type 1 (such as music M) to perform the above process, and then select a piece of music from type 2 (such as music N) to perform the above process, ... In this way, you can get more training data, and for multiple different types of audio information, make the training set richer, and make the subsequent neural network more accurate.

[0065] All the above training data (first training data and second training data) are integrated to serve as a training set for training a neural network. It can be seen that the training set includes multiple training data and sound quality evaluations of the multiple training data. The training data may include audio information and corresponding multiple different equalizer parameters, so that the corresponding sound quality evaluations may represent the evaluations of multiple people on the sound quality of multiple audio information under multiple different equalizer parameters.

[0066] The neural network of the embodiment of the present invention may include two convolutional layers and one fully connected layer, such as Figure 3 As shown. Through multiple iterations, the training of the neural network can be completed.

[0067] It should be noted that Figure 3 The illustration is only an example, for example, a larger number of convolutional layers may be included, for example, a deconvolutional layer, a pooling layer, etc. For example, a loss function may be constructed to determine whether convergence has occurred and thus whether training is completed. The training data may be input into the neural network being trained to obtain a corresponding output, and a loss function may be further constructed based on the corresponding output and the sound quality evaluation corresponding to the training data.

[0068] Exemplarily, in order to reduce the amount of calculation during training, improve the efficiency of training, and shorten the duration of training, the training data may be preprocessed before the neural network is trained. That is, training the neural network may include: preprocessing the training data in the training set to obtain preprocessed data; extracting feature vectors of the preprocessed data; and inputting the feature vectors of the preprocessed data into the neural network for training.

[0069] The preprocessing may include: truncating the training data in the training set according to a predetermined time length to obtain data having multiple predetermined time lengths.

[0070] Subsequently, the feature vectors of each data can be extracted and input into the neural network for training. Specifically, each data of a plurality of predetermined time lengths is subjected to fast Fourier transform (FFT), overlap, and Mel spectrum transform to obtain the feature vector (also called spectrogram) of each data.

[0071] For example, the predetermined time length may be 10 seconds (s), and the training data is cut into segments of 10 seconds. Assuming that a total of 94,860 segments (all 10 seconds) are obtained, each segment is then subjected to a 4196-point FFT transform, a 3072-point overlap, and a 128-point Mel-spectrogram transform to obtain a 427×128 spectrogram.

[0072] In this way, by truncating according to the predetermined time length, the data length is shortened, and thus when the neural network is trained, it can converge faster and shorten the total training time. Figure 4 and Figure 5 The training accuracy and training error are shown respectively. It can be seen that with the increase in the number of iterations, the accuracy gradually approaches 1 and the error gradually approaches 0, that is, convergence is achieved.

[0073] In addition, in order to verify the neural network, the trained neural network can be further verified using the test set, such as Figure 6 As shown, it can be seen that for the test set, an accuracy rate of 70% can be achieved.

[0074] It can be seen that in the process of constructing the training set, the embodiment of the present invention takes into account both objective and subjective methods, and uses deep learning technology for training, thereby obtaining a reliable neural network for evaluating sound quality. The sound quality evaluation obtained by the neural network is not only low-cost and does not require the use of a large number of expert teams, but also the results are accurate and reliable.

[0075] Figure 7 It is a schematic flow chart of the sound quality evaluation method according to an embodiment of the present invention. Figure 7 The process shown includes:

[0076] S110, obtaining the music to be evaluated after the parameters of the equalizer to be tested are adjusted;

[0077] S120, inputting the music to be evaluated into a pre-trained neural network;

[0078] S130, obtaining the evaluation result of the sound quality of the music to be evaluated under the equalizer parameters to be tested output by the neural network.

[0079] For example, if you want to evaluate the sound quality of a certain music at a certain eq parameter, you can adjust the eq parameter of the music to obtain the music to be evaluated in S110. For example, assuming that the equalizer parameter to be tested is eq parameter 1, you can adjust the eq parameter 1 of a certain music to obtain the music to be evaluated.

[0080] Exemplarily, in S120, the feature vector of the music to be evaluated may be input into a pre-trained neural network; or, the feature vector of the music to be evaluated may be first extracted by the neural network.

[0081] Specifically, the music to be evaluated may be subjected to Fourier transform to obtain the feature vector of the music to be evaluated. For example, the music to be evaluated may be subjected to fast Fourier transform (FFT), overlap, and Mel spectrum transform to obtain its feature vector.

[0082] Optionally, the pre-trained neural network includes a feature extraction model, which is used to extract the feature vector of the input music to be evaluated. Specifically, the feature extraction model can perform fast Fourier transform (FFT), overlap, and Mel spectrum transform on the music to be evaluated, and then obtain its feature vector. For example, the feature extraction model can perform a 4196-point fast Fourier transform (FFT), a 3072-point overlap, and a 128-point Mel spectrum transform on the music to be evaluated, and then obtain its feature vector, that is, obtain a 427×128 spectrogram.

[0083] Exemplarily, the pre-trained neural network in S120 may be as described above, and the neural network may include two convolutional layers and one fully connected layer. It is understandable that the complexity of the neural network may be further increased to improve the accuracy of the neural network, such as the number of convolutional layers may be greater, such as a pooling layer may also be included, etc., and the present invention is not limited to this.

[0084] In the process of training the neural network, feature vectors of the training data may be extracted and input into the neural network, and through multiple iterations, a trained neural network in S220 may be obtained.

[0085] Exemplarily, during the training process, the training data may be preprocessed to obtain preprocessed data; feature vectors of the preprocessed data may be extracted; and the feature vectors of the preprocessed data may be input into the neural network for training.

[0086] The preprocessing may include: truncating the training data in the training set according to a predetermined time length to obtain data with multiple predetermined time lengths.

[0087] Subsequently, the feature vectors of each data can be extracted and input into the neural network for training. Specifically, each data of a plurality of predetermined time lengths is subjected to fast Fourier transform (FFT), overlap, and Mel spectrum transform to obtain the feature vector (also called spectrogram) of each data.

[0088] For example, the predetermined time length may be 10 seconds (s), and the training data is cut into segments of 10 seconds. Assuming that a total of 94,860 segments (all 10 seconds) are obtained, each segment is then subjected to a 4196-point FFT transform, a 3072-point overlap, and a 128-point Mel-spectrogram transform to obtain a 427×128 spectrogram.

[0089] In this way, by truncating according to the predetermined time length, the time length of the data is shortened, and thus when the neural network is trained, it can converge faster, thus shortening the total training time.

[0090] Exemplarily, the output of the neural network in S130 may be one of like, medium or dislike. Of course, the output may also be in other forms. It is understandable that the output is related to the neural network. For example, if the sound quality evaluation corresponding to the training data indicates the degree of preference including like, medium and dislike when the neural network is trained, then the output of the neural network is also one of these three. For another example, if the sound quality evaluation corresponding to the training data indicates the degree of preference including more degrees when the neural network is trained, then the output of the neural network is also one of these more degrees.

[0091] It can be seen that the embodiment of the present invention uses a pre-trained neural network to evaluate the sound quality of music, which is simple to operate, does not require the organization of an expert team, and the results are more accurate and reliable.

[0092] Figure 8 It is a schematic block diagram of a sound quality evaluation device according to an embodiment of the present invention. Figure 8 The device 20 shown includes an acquisition module 210 , an input module 220 and an output module 230 .

[0093] The acquisition module 210 may be used to acquire the music to be evaluated after the parameters of the equalizer to be tested are adjusted;

[0094] The input module 220 may be used to input the music to be evaluated into a pre-trained neural network;

[0095] The output module 230 can be used to obtain the evaluation result of the neural network output on the sound quality of the music to be evaluated under the equalizer parameters to be tested.

[0096] Exemplarily, the input module 220 may perform Fourier transform on the music to be evaluated, thereby obtaining a feature vector of the music to be evaluated. For example, the music to be evaluated may be subjected to fast Fourier transform (FFT), overlap, and Mel spectrum transform, thereby obtaining its feature vector.

[0097] For example, the music to be evaluated may be subjected to a 4196-point fast Fourier transform (FFT), a 3072-point overlap, and a 128-point Mel spectrum transform to obtain its feature vector, that is, a 427×128 spectrogram.

[0098] Exemplarily, the output of the output module 230 may be one of like, medium or dislike. Of course, the output may also be in other forms. It is understandable that the output is related to the neural network. For example, if the sound quality evaluation corresponding to the training data indicates the degree of preference including like, medium and dislike when the neural network is trained, then the output of the neural network is also one of these three. For another example, if the sound quality evaluation corresponding to the training data indicates the degree of preference including more degrees when the neural network is trained, then the output of the neural network is also one of these more degrees.

[0099] For example, when constructing a training set, the sound quality evaluation corresponding to the training data may include people's preference for music. As mentioned earlier, 8 different eq parameters can be used to adjust the music, and a 4, 2, and 1 elimination scoring method can be used to determine the degree of preference, where the number of dislike, medium, and like are 4, 3, and 1 respectively.

[0100] Exemplarily, the apparatus 20 may further include a training module, configured to obtain the neural network in S220 through training.

[0101] The training module can be used to: construct a training set; and train the neural network based on the training set. The specific process can be found in the detailed description of training the neural network, which will not be repeated here.

[0102] Figure 8 The device 20 shown can achieve the aforementioned Figure 7 The sound quality evaluation method shown is not repeated here to avoid repetition.

[0103] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0104] In addition, an embodiment of the present invention further provides another sound quality evaluation device, including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the program, the above-mentioned Figure 7 The steps of the sound quality evaluation method are shown.

[0105] like Fig. 9 As shown, the device 30 may include a memory 910 and a processor 920 .

[0106] The memory 910 stores computer program codes for implementing corresponding steps in the sound quality evaluation method according to the embodiment of the present invention.

[0107] The processor 920 is used to run the computer program code stored in the memory 910 to execute the corresponding steps of the sound quality evaluation method according to the embodiment of the present invention.

[0108] Exemplarily, when the computer program code is executed by the processor 920, the following steps are performed: obtaining the music to be evaluated after the parameters of the equalizer to be tested are adjusted; inputting the music to be evaluated into a pre-trained neural network; and obtaining the evaluation result of the neural network output on the sound quality of the music to be evaluated under the parameters of the equalizer to be tested.

[0109] In addition, an embodiment of the present invention further provides a computer storage medium on which a computer program is stored. When the computer program is executed by a processor, the aforementioned Figure 7 The steps of the sound quality evaluation method shown in the figure. For example, the computer storage medium is a computer readable storage medium.

[0110] In one embodiment, the computer program instructions, when executed by a computer or processor, cause the computer or processor to perform the following steps: obtaining the music to be evaluated after the parameters of the equalizer to be tested are adjusted; inputting the music to be evaluated into a pre-trained neural network; and obtaining the evaluation result of the neural network output on the sound quality of the music to be evaluated under the equalizer parameters to be tested.

[0111] Computer storage media may include, for example, a memory card of a smart phone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disk read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. A computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0112] In addition, an embodiment of the present invention further provides a computer program product, which includes instructions. When the instructions are executed by a computer, the computer executes the above Figure 7 The steps of the sound quality evaluation method are shown.

[0113] In one embodiment, when the instruction is executed by a computer, the computer executes the following: obtaining the music to be evaluated after the parameters of the equalizer to be tested are adjusted; inputting the music to be evaluated into a pre-trained neural network; and obtaining the evaluation result of the neural network output on the sound quality of the music to be evaluated under the parameters of the equalizer to be tested.

[0114] It can be seen that the objective evaluation method in the prior art uses instrument testing technical indicators, for example, it can test the frequency response curve of equipment (such as headphones), which ignores people's subjective feelings and the test results are inaccurate. The subjective evaluation method organizes an expert team and the expert team scores, which cannot be used on a large scale and the investment cost is too high. In the process of constructing the training set, the embodiment of the present invention takes into account both objective and subjective methods, and uses deep learning technology for training to obtain a reliable neural network for evaluating sound quality. The sound quality evaluation obtained by the neural network is not only low-cost, without the need to organize a large number of expert team manpower, but also accurate and reliable.

[0115] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of the present invention thereto. Various changes and modifications may be made therein by one of ordinary skill in the art without departing from the scope and spirit of the present invention. All such changes and modifications are intended to be included within the scope of the present invention as required by the appended claims.

[0116] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0117] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0118] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0119] Similarly, it should be understood that in order to streamline the present invention and help understand one or more of the various inventive aspects, in the description of the exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, the method of the present invention should not be interpreted as reflecting the following intention: the claimed invention requires more features than the features explicitly stated in each claim. More specifically, as reflected in the corresponding claims, the inventive point is that the corresponding technical problem can be solved with less than all the features of a single disclosed embodiment. Therefore, the claims following the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present invention.

[0120] It will be understood by those skilled in the art that, except for mutually exclusive features, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this specification may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature that provides the same, equivalent or similar purpose.

[0121] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims, any one of the claimed embodiments may be used in any combination.

[0122] The various component embodiments of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or a digital signal processor (Digital Signal Processing, DSP) can be used in practice to implement some or all functions of some modules in the device according to an embodiment of the present invention. The present invention can also be implemented as a device program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0123] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc., does not indicate any order. These words may be interpreted as names.

[0124] The above is only a specific embodiment of the present invention or an explanation of a specific embodiment, and the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A sound quality evaluation method, characterized in that: include: Obtain the music to be evaluated after the parameters of the equalizer to be tested are adjusted; Inputting the music to be evaluated into a pre-trained neural network; Obtaining the evaluation result of the neural network output on the sound quality of the music to be evaluated under the equalizer parameters to be tested; wherein, The neural network is trained in the following way: Constructing a training set, the training set comprising a plurality of training data and sound quality evaluations corresponding to the plurality of training data, each training data comprising audio information and a corresponding equalizer parameter, and the corresponding sound quality evaluation is an evaluation of the sound quality of the audio information under different equalizer parameters; Training the neural network based on the training set; The construction of the training set includes: for a certain audio information, performing a plurality of different equalizer parameter adjustments, and obtaining a certain person's sound quality evaluation of the audio information after the plurality of different equalizer parameter adjustments, so as to obtain the first training data; for a certain audio information, performing a plurality of different equalizer parameter adjustments, and obtaining a plurality of people's sound quality evaluation of the audio information after the plurality of different equalizer parameter adjustments, so as to obtain the second training data; wherein the sound quality evaluation is used to indicate the degree of preference.

2. The method according to claim 1, characterized in that Inputting the music to be evaluated into a pre-trained neural network includes: Performing Fourier transform on the music to be evaluated, thereby obtaining a feature vector of the music to be evaluated; The feature vector of the music to be evaluated is input into the pre-trained neural network.

3. The method according to claim 1, characterized in that The training of the neural network based on the training set includes: Preprocessing the training data in the training set to obtain preprocessed data; Extracting a feature vector of the preprocessed data; The feature vector is input into the neural network for training.

4. The method according to claim 3, characterized in that The pre-processing comprises: The training data in the training set is truncated according to a predetermined time length to obtain data with multiple predetermined time lengths.

5. The method according to claim 3, characterized in that: The step of extracting a feature vector of the preprocessed data comprises: The preprocessed data are subjected to fast Fourier transform, overlap, and Mel spectrum transform to obtain a feature vector of each data.

6. A sound quality evaluation device, characterized in that: include: An acquisition module, used for acquiring the music to be evaluated after the parameters of the equalizer to be tested are adjusted; An input module, used for inputting the music to be evaluated into a pre-trained neural network; An output module, used for obtaining the evaluation result of the neural network output on the sound quality of the music to be evaluated under the equalizer parameters to be tested; The neural network is trained in the following way: Constructing a training set, the training set comprising a plurality of training data and sound quality evaluations corresponding to the plurality of training data, each training data comprising audio information and a corresponding equalizer parameter, and the corresponding sound quality evaluation is an evaluation of the sound quality of the audio information under different equalizer parameters; Training the neural network based on the training set; The construction of the training set includes: for a certain audio information, performing a plurality of different equalizer parameter adjustments, and obtaining a certain person's sound quality evaluation of the audio information after the plurality of different equalizer parameter adjustments, so as to obtain the first training data; for a certain audio information, performing a plurality of different equalizer parameter adjustments, and obtaining a plurality of people's sound quality evaluation of the audio information after the plurality of different equalizer parameter adjustments, so as to obtain the second training data; wherein the sound quality evaluation is used to indicate the degree of preference.

7. A sound quality evaluation device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images