Speech quality assessment method and apparatus, storage medium, and electronic device

Abstract

A speech quality assessment method and apparatus, a computer-readable storage medium, and an electronic device, relating to the technical field of computers. The method comprises: acquiring a corpus file and a corresponding recording file; on the basis of a trained assessment model, respectively extracting a corpus feature corresponding to the corpus file, and a recording feature corresponding to the recording file; and fusing the corpus feature and the recording feature to obtain a fused feature, and fitting the fused feature to obtain an assessment result.

Description

Voice quality assessment method and device, storage medium, and electronic device

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Chinese patent application number 202310877995.9, filed on July 17, 2023, entitled “Speech quality assessment method and device, storage medium, electronic device”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present disclosure relates to the field of computer technology, and in particular to a speech quality assessment method, a speech quality assessment device, a computer-readable storage medium, and an electronic device. Background Art

[0004] In order to improve the call quality during mobile communication, the call quality during mobile communication can be evaluated.

[0005] Related technologies can use MOS (Mean Opinion Score) for prediction. Specifically, specialized instruments are used to process the degraded signal after wireless network transmission, transform it, and compare it with the original signal. This is then mapped to a prediction of the subjective mean opinion score using a cognitive model.

[0006] The above method uses two communication terminals to play and record audio, comparing the lossy waveform after network transmission with the original waveform. This method uses a subjective model that cannot be directly correlated with user perception. Furthermore, it may result in a low MOS score and inaccurate evaluation.

[0007] Summary of the Invention

[0008] According to a first aspect of the present disclosure, a speech quality assessment method is provided, comprising: obtaining a corpus file and a corresponding recording file; extracting corpus features corresponding to the corpus file and extracting recording features corresponding to the recording file based on a trained assessment model; fusing the corpus features and the recording features to obtain fused features, and fitting the fused features to obtain an assessment result.

[0009] According to a second aspect of the present disclosure, a speech quality assessment device is provided, comprising: a file acquisition module for acquiring a corpus file and a corresponding recording file; extracting corpus features corresponding to the corpus file and recording features corresponding to the recording file based on a trained assessment model; fusing the corpus features and recording features to obtain fused features, and fitting the fused features to obtain an assessment result.

[0010] According to a third aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any one of the above-mentioned speech quality assessment methods is implemented.

[0011] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform any one of the above-mentioned speech quality assessment methods by executing the executable instructions. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] FIG1 schematically shows a system architecture diagram for speech quality assessment according to an embodiment of the present disclosure.

[0013] FIG2 schematically shows a flow chart of a method for evaluating speech quality according to an embodiment of the present disclosure.

[0014] FIG3 schematically shows a flow chart of pre-processing an original recording file according to an embodiment of the present disclosure.

[0015] FIG4 schematically shows a specific flow chart of the transmission between corpus files and recording files according to an embodiment of the present disclosure.

[0016] FIG5 schematically shows a flow chart of evaluation based on an evaluation model according to an embodiment of the present disclosure.

[0017] FIG6 schematically shows a flow chart of evaluation model training according to an embodiment of the present disclosure.

[0018] FIG7 schematically illustrates a schematic diagram of a specific process of training an evaluation model according to an embodiment of the present disclosure.

[0019] FIG8 schematically illustrates a flow chart of quality assessment according to a trained assessment model in an embodiment of the present disclosure.

[0020] FIG9 schematically shows a block diagram of a speech quality assessment apparatus according to an embodiment of the present disclosure.

[0021] FIG10 schematically shows a block diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0022] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present disclosure will be more comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or that other methods, components, devices, steps, etc. may be employed. In other cases, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0023] In addition, the accompanying drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0024] In related technologies, voice quality assessment can be performed using a MOS-based approach. Specifically, a professional instrument processes the degraded signal after wireless network transmission through level adjustment, filtering, and other processing steps. The transformed signal is then compared with the original signal and mapped to a predicted subjective mean opinion score using a cognitive model.

[0025] The above method may have the following issues: Playback and recording tests are performed using two communication terminals, and the difference between the lossy waveform after network transmission and the original waveform is compared. This uses a subjective model that cannot be directly correlated with user perception. If the recording file deviates from the corpus file by more than two seconds, the MOS score will be low. As mobile terminals improve their noise reduction processing capabilities, the existing voice evaluation method is prone to evaluation inaccuracies.

[0026] To address the aforementioned technical issues, the disclosed embodiments provide a voice quality assessment method applicable to mobile communication network-based communication scenarios. By introducing a classification neural network into the algorithm model and conducting large-scale training of the scoring network based on specific environments and factors, voice quality assessment is performed through corpus file processing, voice recording, assessment model training, and voice evaluation. This method makes the quality assessment results of voice calls initiated over mobile network environments such as 4G, 5G, and WiFi more realistic.

[0027] Figure 1 schematically illustrates an application scenario for mobile network-based communication. Referring to Figure 1 , communication terminals 101 and 102 are conducting voice communication over the mobile network. During the communication process, either terminal can be used as the transmitter and the other as the receiver. This example uses terminal 101 as the transmitter and terminal 102 as the receiver.

[0028] In some embodiments, a corpus file can be played at a transmitting end and transmitted to a receiving end via a mobile network, so that a recording file can be output at the receiving end. The recording file can be lossy. Furthermore, the corpus file and the recording file can be input into a trained evaluation model, which processes the corpus file and the recording file and outputs an evaluation result, thereby determining the voice quality based on the evaluation result.

[0029] In this disclosed embodiment, a trained evaluation model extracts corpus features from corpus files and audio features from audio files. These features are then fused to create a fused feature, which is then used to generate an evaluation result. Because the evaluation process can be completed on a cloud platform, there's no need for an external mobile terminal device, reducing testing costs. The model can also adapt to multiple speech coding rates.

[0030] Next, with reference to FIG2 , each step of the speech quality assessment method in the embodiment of the present disclosure is described in detail.

[0031] In step S210, a corpus file and a corresponding recording file are obtained.

[0032] In the embodiment of the present disclosure, the corpus file may be a corpus file obtained by processing the original corpus file. The original corpus file may be composed of corpus files of multiple types and languages. In order to improve accuracy, the original corpus file may be processed. For example, based on the ITU-P863 standard, the original corpus file may be band-pass filtered and volume-uniformed. The processed original corpus file may further be used as a corpus file for intelligent speech quality assessment. Band-pass filtering is used to filter the original corpus file, and volume-uniformed processing refers to processing the volume of the original corpus file to the same volume, and the same volume may be, for example, -20db.

[0033] The corpus file can be the corpus file played by the transmitter, and the recording file can be the lossy recording file output by the receiver. Its content can correspond to the corpus file. Voice communication between the transmitter and receiver can be conducted over a mobile communication network. The transmission method between the transmitter and receiver can be any of 4G, 5G networks, or WiFi. The transmitter transmits the corpus file to the receiver via the mobile communication network, so that the receiver outputs the lossy recording file.

[0034] In the disclosed embodiments, a pre-recorded voice file can be played at the transmitter through the instrument and used as a corpus file. The transmitter can be a terminal on the same carrier's network or a terminal on a different carrier's network. The transmitter can be a mobile phone, CPE (Customer Premise Equipment), tablet computer, or any other device capable of providing voice services on the mobile network. Specifically, the corpus file played can be in any language, such as Chinese, English, or French; the corpus file can include voice files of male or female speakers, and can include voices of all age groups. The corpus file played can originate from any media, including movies, television programs, radio programs, and internet audio and video. During the playback of the corpus file, playback information for the corpus file can be set. This playback information can include one or more of the following: the volume of the voice file, the duration of the voice file, and the playback mode. For example, the volume of the voice file can be set to 80%; the duration of the corpus file can be set to 8 seconds; and the playback mode can be set to one or more of the following: continuous playback, intermittent playback, and repeated playback.

[0035] Exemplarily, as shown in reference figure 3, the original corpus file can be preprocessed, and the preprocessed original corpus file can be processed accordingly based on whether the volume of the preprocessed original corpus file meets the volume requirement, and when the volume meets the volume requirement, the preprocessed original corpus file can be processed accordingly. Among them, the preprocessing can be a filtering process, and the filtering process can specifically be a bandpass filtering or other filtering methods. Based on whether the volume of the preprocessed original corpus file meets the volume requirement, the corresponding processing of the preprocessed original corpus file specifically includes: if the volume of the preprocessed original corpus file is greater than or equal to the volume threshold, the preprocessed original corpus file is used as the corpus file to execute the subsequent speech quality evaluation process. If the volume of the preprocessed original corpus file is less than the volume threshold, the MOS score can be directly output as 1. Among them, the volume threshold is used to measure whether the volume of the corpus file is within the normal range, and the volume threshold can be -80db.

[0036] During the communication between the sending end and the receiving end, while the sending end plays the corpus file, the receiving end can record the received corpus file and use the received corpus file as the recording file. Further, a speech quality assessment process can be implemented based on the received recording file. Referring to Figure 4, the corpus file, that is, the original standard corpus file, can be transmitted via a mobile network, and the transmitted corpus file can be used as the recording file. The recording file can be a lossy recording file. The transmission method can be any one of a 4G network, a 5G network, or a WiFi method.

[0037] Next, in step S220, the corpus features corresponding to the corpus file and the recording features corresponding to the recording file are extracted based on the trained evaluation model.

[0038] In the disclosed embodiment, the corpus file and the recording file can be input into the trained evaluation model to evaluate the corpus file and the recording file using the trained evaluation model to obtain an evaluation score as the evaluation result. Referring to FIG5 , the trained evaluation model can be obtained according to the evaluation model training method.

[0039] It should be noted that to improve the accuracy of the evaluation model, the model can be trained to obtain a trained evaluation model, which can then be used to predict the speech quality of corpus files and recording files. To improve the accuracy of the evaluation model, during the training process, the default classification is greater than or equal to 4,000 categories; the training process requires 1 million to 4 million POLQA corpus data; the MOS scores of the corpus data must cover the range of [1 point, 5 points] and be evenly distributed. In addition, to ensure that the evaluation results are more comprehensive across the entire test data domain, corpus sample data with typical characteristics can be added to enrich the classification neural network, making the evaluation process more realistic and accurate.

[0040] FIG6 schematically shows a flow chart of performing evaluation model training. Referring to FIG6 , the flow chart mainly includes the following steps:

[0041] In step S610, the sample corpus file and the sample recording file are evaluated based on a reference evaluation method to obtain a reference evaluation result;

[0042] In step S620, the sample corpus file and the sample recording file are evaluated based on the evaluation model to obtain a prediction evaluation result;

[0043] In step S630, the reference evaluation result and the predicted evaluation result are analyzed, and the evaluation model is trained according to the analysis result to obtain a trained evaluation model.

[0044] In the disclosed embodiments, the reference evaluation method may be the POLQA evaluation method. The POLQA algorithm is a new generation voice quality assessment standard applicable to voice quality assessment in fixed-line, mobile, and IP networks. POLQA has been established as Recommended Specification P.863 and can be used for voice quality assessment in HD voice, 3G, 4G / VoLTE, and 5G networks. For example, a sample corpus file and a sample recording file can both be used as inputs to the POLQA evaluation method to perform a POLQA evaluation, thereby obtaining a POLQA evaluation output as a reference evaluation result. Specifically, the signals of the sample corpus file and the sample recording file are filtered, time-aligned, and sample rate estimated. These signals are then input into a perceptual model to obtain internal representations of the signals of the sample corpus file and the sample recording file. The difference between these internal representations is then input into a cognitive model to obtain a POLQA score, which is then mapped to a MOS-LQO score as a reference score. The reference evaluation result may be a standard score obtained by evaluating the sample corpus file and the sample recording file, for example, 3.0 or other numerical values.

[0045] For a sample recording file, if the reference evaluation result given by the standard library is 3.0, then some parameters corresponding to the reference evaluation result can be extracted from the standard library according to the reference evaluation result, and features such as frequency and sound can be extracted. An evaluation model is constructed based on the frequency features and sound features, and the score of the sample recording file is predicted based on the evaluation model.

[0046] Next, the sample corpus file and the sample recording file can be evaluated by the evaluation model to obtain a predicted evaluation result. For example, the sample corpus file and the sample recording file can be input into the evaluation model, and the sample corpus file and the sample recording file can be pre-emphasized, framed, and sliced ​​by the evaluation model to obtain the corresponding plum spectrum; the plum spectrum is convolved to obtain the corpus features corresponding to the sample corpus file and the recording features corresponding to the sample recording file. The corpus features and the recording features are spliced ​​to obtain the spliced ​​features; the spliced ​​features are fully connected to obtain the predicted evaluation result. The predicted evaluation result can be a prediction score based on the trained evaluation model, which can be the same as or different from the reference evaluation result, and is specifically calculated according to the model training.

[0047] Furthermore, an analysis result can be obtained based on the reference evaluation result and the predicted evaluation result, and the evaluation model can be trained based on the analysis result. For example, referring to FIG7, the reference evaluation result and the predicted evaluation result can be measured based on the loss function to obtain difference data between the reference evaluation result and the predicted evaluation result. After obtaining the difference data, the difference data can be subjected to DIFF processing for comparative analysis to obtain target difference data with differences.

[0048] On this basis, the derivative change rate can be calculated based on the target difference data to smooth the target difference data to obtain smoothed difference data. The model parameters of the evaluation model can then be adjusted based on the smoothed difference data, i.e., the parameters of the evaluation model can be adjusted based on the smoothed difference data, i.e., Net Change processing can be performed. Furthermore, an adjusted evaluation model can be determined based on the adjusted model parameters, and the sample corpus file and sample recording file can be processed based on the adjusted evaluation model to obtain a predicted evaluation result. The evaluation model can then be iteratively trained based on the predicted evaluation result until a training end condition is met, thereby obtaining a trained evaluation model. The training end condition can be that the difference between the output predicted evaluation result and the reference evaluation result is less than a difference threshold, or that the number of training cycles reaches a number threshold. The difference threshold can be a relatively small value, such as 0.0001 or other values, and can be determined based on actual needs. The number threshold can also be determined based on actual needs.

[0049] It should be noted that when updating the model parameters of the evaluation model, the model parameters can be updated based on the learning rate. For different iterative training processes, since the model parameters of the evaluation model can be updated with each iteration, the predicted evaluation results obtained based on the evaluation models with different parameters may also be different. This may result in different difference data during each training process. Therefore, the learning rate used in each training process can be the same or different. The learning rate is used to determine whether the objective function can converge to a local minimum and when it converges to the minimum. The learning rate can be increased or decreased according to the number of training times, and is specifically determined based on the actual difference data.

[0050] During the evaluation process, as shown in FIG7 , since the trained evaluation model may include MFCC processing, CNN processing, CAT processing, and DNN processing, the corpus file and the recording file may be subjected to MFCC processing and CNN processing, respectively.

[0051] First, we will explain the MFCC processing of the corpus file. The MFCC processing process can be a Mel-frequency cepstrum operation. MFCC represents the short-term power spectrum of speech and is formed by performing a linear cosine transform on the nonlinear Mel-frequency scale on the logarithmic power spectrum of the speech in the corpus file. During the MFCC processing process, key operations include pre-emphasis, framing, and slicing the signal. Based on this, a Mel-frequency cepstrum corresponding to the corpus file and a Mel-frequency cepstrum corresponding to the recording file can be obtained. The result of the MFCC processing process, i.e., the graph represented by the Mel-frequency cepstrum, is also a two-dimensional image, i.e., two-dimensional data.

[0052] Next, the CNN processing process can be performed on the two-dimensional data corresponding to the plum spectrum graph to perform convolution processing to obtain the corpus features corresponding to the corpus file. The corpus features can be one-dimensional features. For example, the two-dimensional data can be convolved and a product-sum process can be implemented through a cross-correlation operation to process the two-dimensional data into one-dimensional features, that is, to obtain the corpus features corresponding to the corpus file.

[0053] The process for processing recording files is the same as that for processing corpus files. For example, the recording file can be pre-emphasized, framed, and sliced ​​to obtain a plum spectrum corresponding to the recording file. This plum spectrum corresponding to the recording file can then be convolved to convert the two-dimensional data represented by the plum spectrum corresponding to the recording file into a one-dimensional feature, which is then determined as the recording feature corresponding to the recording file.

[0054] Next, referring to FIG. 2 , in step S230 , the corpus features and the recording features are fused to obtain fused features, and the fused features are fitted to obtain an evaluation result.

[0055] In the embodiment of the present disclosure, after obtaining the trained evaluation model, the corpus features and the recording features obtained by feature extraction can be fused based on the trained evaluation model to obtain fused features. The fusion here can be a splicing CAT process, and the fused features obtained are splicing features. Furthermore, the obtained splicing features can be fully connected to achieve fitting of the splicing features, thereby obtaining an evaluation result. Exemplarily, the obtained fused features can be input into the DNN layer for full connection processing, so that the trained evaluation model outputs an evaluation result. Among them, the full connection processing can be achieved by a convolution operation.

[0056] The evaluation result can be a MOS score. The MOS (Mean Opinion Score) score uses sound waveform attenuation scoring technology. Playback and recording tests are performed on two mobile phones to compare the difference between the lossy waveform and the original waveform after network transmission. A subjective model is used and cannot be directly associated with user perception. The MOS score can be obtained by predicting the score of the difference between the corpus file and the recording file, indicating the quality of the recording file corresponding to the corpus file, thereby characterizing the voice quality during voice transmission based on the mobile network. In addition, the MOS score is positively correlated with quality, that is, the higher the MOS score, the better the quality.

[0057] The technical solutions in the disclosed embodiments can switch corpus files at will without affecting the evaluation process, making them more suitable for various practical applications and adaptable to multiple speech coding rates. They are versatile and offer greater authenticity. Because voice quality prediction is automatically achieved through a trained evaluation model, low MOS scores due to factors such as the terminal sound card (e.g., noise reduction) are reduced, improving the accuracy of the scores and the precision of the voice quality assessment. Because the trained evaluation model incorporates a classification neural network, it can be enriched and improved during the training process, making the trained evaluation model more accurate and enabling it to address various new issues, thereby improving the comprehensiveness and accuracy of the estimates. Furthermore, the voice quality assessment process can be completed for recorded files on a cloud platform based on the trained evaluation model, eliminating the need for external mobile phone terminal devices, reducing testing costs, and improving processing efficiency. This can enhance operators' ability to assess the perception of voice services such as VoNR and VoLTE on mobile networks, allowing them to directly understand users' voice service perceptions. This minimizes MOS assessment inaccuracies caused by factors such as terminal equipment and test personnel. The quality of modern wireless communication networks is generally good. This patent can promote the refinement or redefinition of the MOS scoring standard to achieve more focused or practical voice quality evaluation.

[0058] The present disclosure also provides a speech quality assessment device. Referring to FIG9 , the speech quality assessment device 900 mainly includes the following modules:

[0059] The file acquisition module 901 is used to acquire the corpus file and the corresponding recording file;

[0060] A feature extraction module 902 is configured to extract corpus features corresponding to the corpus file and recording features corresponding to the recording file based on the trained evaluation model;

[0061] The evaluation prediction module 903 is used to fuse the corpus features and the recording features to obtain fused features, and to fit the fused features to obtain an evaluation result.

[0062] In an exemplary embodiment of the present disclosure, the extracting of corpus features corresponding to the corpus file and the extracting of recording features corresponding to the recording file based on the trained evaluation model respectively include: pre-emphasizing, framing and slicing the corpus file and the recording file respectively to obtain corresponding plum spectrum graphs; performing convolution processing on the plum spectrum graph to obtain corpus features corresponding to the corpus file and recording features corresponding to the recording file.

[0063] In an exemplary embodiment of the present disclosure, the fusing of the corpus features and the recording features to obtain fused features, and fitting the fused features to obtain evaluation results, include: splicing the corpus features and the recording features to obtain splicing features; and performing full connection processing on the splicing features to obtain the evaluation results.

[0064] In an exemplary embodiment of the present disclosure, the method further includes: evaluating the sample corpus file and the sample recording file based on a reference evaluation method to obtain a reference evaluation result; evaluating the sample corpus file and the sample recording file based on an evaluation model to obtain a predicted evaluation result; analyzing the reference evaluation result and the predicted evaluation result in combination, and training the evaluation model according to the analysis result to obtain a trained evaluation model.

[0065] In an exemplary embodiment of the present disclosure, the reference evaluation result and the predicted evaluation result are analyzed in combination, and the evaluation model is trained according to the analysis result to obtain a trained evaluation model, including: measuring the reference evaluation result and the predicted evaluation result based on a loss function to obtain difference data between the reference evaluation result and the predicted evaluation result; and training the evaluation model based on the difference data to obtain a trained evaluation model.

[0066] In an exemplary embodiment of the present disclosure, the evaluation model is trained based on the difference data to obtain a trained evaluation model, including: performing comparative analysis on the difference data, and determining the derivative change rate corresponding to the difference data to smooth the difference data; and iteratively adjusting the model parameters of the evaluation model based on the smoothed difference data until the training end condition is met to obtain the trained evaluation model.

[0067] In an exemplary embodiment of the present disclosure, the model parameters of the evaluation model are iteratively adjusted based on the smoothed difference data until the training end condition is met to obtain the trained evaluation model, including: according to the learning rate, the model parameters of the evaluation model are iteratively adjusted based on the smoothed difference data until the difference between the output predicted evaluation result and the reference evaluation result is less than the difference threshold to obtain the trained evaluation model.

[0068] It should be noted that the specific details of each module in the above-mentioned speech quality assessment device have been described in detail in the corresponding speech quality assessment method, and therefore will not be repeated here.

[0069] It should be noted that although several modules or units of the device for action execution are mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0070] Furthermore, although the steps of the method of the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that the steps must be performed in this particular order, or that all steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0071] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0072] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods, or program products. Therefore, various aspects of the present disclosure may be implemented in the following forms: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits," "modules," or "systems."

[0073] The electronic device 1000 according to this embodiment of the present disclosure is described below with reference to Figure 10. The electronic device 1000 shown in Figure 10 is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0074] As shown in FIG10 , electronic device 1000 is implemented as a general-purpose computing device. Components of electronic device 1000 may include, but are not limited to, the aforementioned at least one processing unit 1010, the aforementioned at least one storage unit 1020, a bus 1030 connecting various system components (including storage unit 1020 and processing unit 1010), and a display unit 1040.

[0075] The storage unit stores program code, which can be executed by the processing unit 1010, so that the processing unit 1010 performs the steps described in the "Exemplary Method" section above according to various exemplary embodiments of the present disclosure. For example, the processing unit 1010 can perform the steps shown in Figure 2.

[0076] The storage unit 1020 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 10201 and / or a cache memory unit 10202 , and may further include a read-only memory unit (ROM) 10203 .

[0077] The storage unit 1020 may also include a program / utility 10204 having a set (at least one) of program modules 10205, such program modules 10205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0078] Bus 1030 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0079] The electronic device 1000 can also communicate with one or more external devices 1100 (e.g., a keyboard, a pointing device, a Bluetooth device, etc.), one or more devices that enable a user to interact with the electronic device 1000, and / or any device that enables the electronic device 1000 to communicate with one or more other computing devices (e.g., a router, a modem, etc.). Such communication can occur via an input / output (I / O) interface 1050. Furthermore, the electronic device 1000 can also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 1060. As shown, the network adapter 1060 communicates with other modules of the electronic device 1000 via a bus 1030. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 1000, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0080] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or an electronic device, etc.) to execute the method according to the embodiments of the present disclosure.

[0081] In exemplary embodiments of the present disclosure, a computer-readable storage medium is also provided, on which is stored a program product capable of implementing the aforementioned methods of this specification. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product comprising program code. When the program product is executed on a terminal device, the program code is configured to cause the terminal device to execute the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of the present disclosure.

[0082] According to an embodiment of the present disclosure, a program product for implementing the above-mentioned method can be a portable compact disc read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0083] The program product may be implemented in any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0084] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0085] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0086] The program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0087] Furthermore, the figures above are merely illustrative of the processes included in the methods according to exemplary embodiments of the present disclosure and are not intended to be limiting. It is readily understood that the processes illustrated in the figures above do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0088] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow from the general principles of the present disclosure and include common knowledge or customary techniques in the art not invented herein. The specification and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the claims.

Claims

1. A method for evaluating speech quality, comprising: Get the corpus file and the corresponding recording file; Extracting the corpus features corresponding to the corpus file and extracting the recording features corresponding to the recording file based on the trained evaluation model respectively; The corpus features and the recording features are fused to obtain fused features, and the fused features are fitted to obtain evaluation results.

2. The method for evaluating speech quality according to claim 1, wherein: The extracting the corpus features corresponding to the corpus file and the extracting the recording features corresponding to the recording file based on the trained evaluation model respectively include: Pre-emphasize, frame and slice the corpus file and the recording file respectively to obtain corresponding plum spectrum diagrams; The plum spectrum graph is convoluted to obtain corpus features corresponding to the corpus file and recording features corresponding to the recording file.

3. The method for evaluating speech quality according to claim 1, wherein: The fusing of the corpus feature and the recording feature to obtain a fusion feature, and fitting the fusion feature to obtain an evaluation result, includes: Concatenating the corpus feature and the recording feature to obtain a concatenated feature; Fully connect the splicing features to obtain the evaluation result.

4. The method for evaluating speech quality according to claim 1, wherein: The method further comprises: Evaluate the sample corpus files and sample recording files based on the reference evaluation method to obtain reference evaluation results; Evaluate the sample corpus files and sample recording files based on the evaluation model to obtain the predicted evaluation results; The reference evaluation result and the predicted evaluation result are analyzed in combination, and the evaluation model is trained according to the analysis result to obtain a trained evaluation model.

5. The method for evaluating speech quality according to claim 4, wherein: The combining the reference evaluation result and the predicted evaluation result for analysis, and training the evaluation model according to the analysis result to obtain a trained evaluation model, includes: Measuring the reference evaluation result and the predicted evaluation result based on a loss function to obtain difference data between the reference evaluation result and the predicted evaluation result; The evaluation model is trained based on the difference data to obtain a trained evaluation model.

6. The method for evaluating speech quality according to claim 5, wherein: The step of training the evaluation model based on the difference data to obtain a trained evaluation model includes: Comparatively analyzing the difference data, and determining a derivative change rate corresponding to the difference data, so as to smooth the difference data; The model parameters of the evaluation model are iteratively adjusted based on the smoothed difference data until a training end condition is met to obtain the trained evaluation model.

7. The method for evaluating speech quality according to claim 6, wherein: The iterative adjustment of the model parameters of the evaluation model based on the smoothed difference data until the training end condition is met to obtain the trained evaluation model includes: According to the learning rate, the model parameters of the evaluation model are iteratively adjusted based on the smoothed difference data until the difference between the output predicted evaluation result and the reference evaluation result is less than the difference threshold, so as to obtain the trained evaluation model.

8. A speech quality assessment device, comprising: File acquisition module, used to obtain corpus files and corresponding recording files; A feature extraction module, used to extract the corpus features corresponding to the corpus file and the recording features corresponding to the recording file based on the trained evaluation model; The evaluation prediction module is used to fuse the corpus features and the recording features to obtain fused features, and to fit the fused features to obtain evaluation results.

9. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the speech quality assessment method according to any one of claims 1 to 7 is implemented.

10. An electronic device comprising: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the speech quality assessment method according to any one of claims 1 to 7 by executing the executable instructions.

Images