Speech quality assessment method based on classification and feature extraction

Abstract

The present application relates to a method for speech quality assessment based on classification and feature extraction. The method comprises: obtaining speech data to be assessed, and performing Fourier transformation on each frame of speech in the speech data, and then correspondingly counting the amplitude information thereof to obtain the amplitude information corresponding to each frame of speech, classifying each frame of speech according to the amplitude information corresponding to each frame of speech and preset parameters, and performing feature extraction on each frame of speech whose classification results are the second category and the third category, and evaluating the corresponding frame of speech according to the extracted features to obtain the assessment results of each frame of speech in the second category and the third category, and performing mean calculation according to the assessment results of each frame of speech in the second category and the third category to obtain the quality assessment result of the speech data to be assessed. The use of this method can fully consider the differences between speech signals, and adopt different quality assessment methods for different speech signals, with good evaluation effect and low algorithm complexity, and has good practical value.

Description

Technical Field

[0001] The present application relates to the technical field of speech quality evaluation, and in particular to a speech quality evaluation method based on classification and feature extraction. Background Art

[0002] In recent years, with the rapid development of digital network and VLSI technology, various audio and video processing technologies and audio and video transmission technologies have emerged one after another. Since the subjective feelings of communication users and consumers for audio and video ultimately depend on the audio and video quality, voice and video quality evaluation has become an increasingly important research topic.

[0003] Speech quality evaluation can be divided into two categories based on the evaluation subject: subjective evaluation and objective evaluation. Subjective quality evaluation can accurately reflect speech quality, but the implementation steps are complicated and the real-time performance is poor, so it is not suitable for quality evaluation in real-time voice communication. Objective quality evaluation methods are more applicable to network voice services and are currently a hot topic in the field of speech quality evaluation research.

[0004] According to whether the original signal is needed, objective evaluation methods are divided into full reference and no reference. The full reference evaluation method judges the quality of speech by comparing the difference between the original signal and the distorted signal. At present, the full reference objective evaluation algorithm for speech quality recommended by the International Telecommunication Union is based on the Perceptual Evaluation of Speech Quality (PESQ) model. Compared with previous algorithms, the PESQ model has better recognition capabilities for communication delay, environmental noise and errors, but the algorithm is more complex and is not suitable for real-time evaluation of speech quality, nor is it suitable for implementation on micro systems with limited computing power. The no reference evaluation method only evaluates speech quality based on distorted signals or statistical network parameters, which can be applied to different language environments. The G.107 standard formulated by ITU-T uses the average network packet loss rate to predict network speech quality. This method does not require original speech information. However, the degree of network speech quality degradation is not only related to the number of lost packets, but also to the content characteristics of the lost packets. Packet loss with different content characteristics has different effects on speech quality. This type of method does not fully consider these factors, so the accuracy is not high. Summary of the invention

[0005] Based on this, it is necessary to provide a speech quality assessment method based on classification and feature extraction to address the above technical issues.

[0006] 1. A speech quality assessment method based on classification and feature extraction, characterized in that the method comprises:

[0007] Acquire speech data to be evaluated, and perform Fourier transform on each frame of speech in the speech data and then count the amplitude information thereof accordingly, so as to obtain amplitude information corresponding to each frame of speech;

[0008] Classifying each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include a first category, a second category and a third category, extracting features of each speech frame whose classification results are the second category and the third category, and evaluating the corresponding speech frame according to the extracted features to obtain evaluation results of each speech frame of the second category and the third category;

[0009] The quality assessment result of the speech data to be assessed is obtained by performing mean calculation based on the assessment results of each speech frame of the second category and the third category.

[0010] In one embodiment, the classification of each frame of speech according to the amplitude information corresponding to each frame of speech and the preset parameters specifically includes:

[0011] When the amplitude information corresponding to a frame of speech is less than the preset parameter, the frame of speech is classified into the first category;

[0012] When the amplitude information corresponding to a frame of speech is greater than or equal to the preset parameter and less than 5 times the preset parameter, the frame of speech is classified into the second category;

[0013] When the amplitude information corresponding to a frame of speech is greater than or equal to 5 times the preset parameter, the frame of speech is classified into the third category.

[0014] In one embodiment, extracting features from each speech frame classified as the second category, and evaluating the corresponding speech frame according to the extracted features specifically includes:

[0015] A bandpass filter is used to filter a frame of speech to obtain a filtered speech signal;

[0016] Performing feature extraction based on the filtered speech signal to obtain a first feature representing human auditory perception;

[0017] The first features are respectively mapped to a quality space to obtain an evaluation result of the speech frame.

[0018] In one embodiment, the first feature extraction is expressed as:

[0019]

[0020] In the above formula, y2 i (n) is the filtered speech signal, λ1 and λ2 are preset parameters, and their value range is between [0.1]. i This is the first feature.

[0021] In one embodiment, the first features are respectively mapped to a quality space to obtain an evaluation result of the frame of speech, using the following formula:

[0022]

[0023] In the above formula, MOS i is the evaluation result of each frame of speech.

[0024] In one embodiment, the step of extracting features from each speech frame classified as the third category and evaluating the corresponding speech frame according to the extracted features specifically includes:

[0025] A frame of speech is filtered using a GammaTone filter to obtain a filtered speech signal;

[0026] Performing feature extraction based on the filtered speech signal to obtain two second features and a third feature representing human auditory perception;

[0027] An aggregation factor is calculated according to the second feature and the third feature, and an evaluation result of the frame of speech is obtained by performing calculation according to the aggregation factor.

[0028] In one embodiment, the second feature extraction is expressed as:

[0029] F2 i =λ3×exp(-1 / y3 i (n))

[0030] In the above formula, y3 i (n) is the filtered speech signal, λ3 is a preset parameter whose value range is between [1, 10], F2 i is the second feature;

[0031] The third feature extraction is expressed as:

[0032]

[0033] In the above formula, y3 i (n) is the filtered speech signal, λ4 and λ5 are preset parameters, whose value range is between [10, 20], and F3 i This is the third feature.

[0034] In one embodiment, the aggregation factor is calculated according to the second feature and the third feature, wherein the following formula is used for calculation:

[0035]

[0036] In the above formula, μ1 iRepresents the aggregation factor.

[0037] In one embodiment, the evaluation result of the frame speech is obtained by calculating according to the aggregation factor, using the following formula:

[0038]

[0039] In the above formula, MOS i is the evaluation result of each frame of speech.

[0040] In one embodiment, the duration of the speech data to be evaluated is less than 10 seconds.

[0041] A speech quality assessment device based on classification and feature extraction, the device comprising:

[0042] A voice data acquisition module is used to acquire the voice data to be evaluated, and to perform Fourier transform on each frame of voice in the voice data and then to count the amplitude information thereof, so as to obtain the amplitude information corresponding to each frame of voice;

[0043] The speech quality assessment module for each frame is used to classify each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include the first category, the second category and the third category, extract features of each speech frame whose classification results are the second category and the third category, and evaluate the corresponding speech frame according to the extracted features to obtain the evaluation results of each speech frame of the second category and the third category;

[0044] The quality assessment result obtaining module is used to perform mean calculation based on the assessment results of each speech frame of the second category and the third category to obtain the quality assessment result of the speech data to be assessed.

[0045] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0046] Acquire speech data to be evaluated, and perform Fourier transform on each frame of speech in the speech data and then count the amplitude information thereof accordingly, so as to obtain amplitude information corresponding to each frame of speech;

[0047] Classifying each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include a first category, a second category and a third category, extracting features of each speech frame whose classification results are the second category and the third category, and evaluating the corresponding speech frame according to the extracted features to obtain evaluation results of each speech frame of the second category and the third category;

[0048] The quality assessment result of the speech data to be assessed is obtained by performing mean calculation based on the assessment results of each speech frame of the second category and the third category.

[0049] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0050] Acquire speech data to be evaluated, and perform Fourier transform on each frame of speech in the speech data and then count the amplitude information thereof accordingly, so as to obtain amplitude information corresponding to each frame of speech;

[0051] Classifying each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include a first category, a second category and a third category, extracting features of each speech frame whose classification results are the second category and the third category, and evaluating the corresponding speech frame according to the extracted features to obtain evaluation results of each speech frame of the second category and the third category;

[0052] The quality assessment result of the speech data to be assessed is obtained by performing mean calculation based on the assessment results of each speech frame of the second category and the third category.

[0053] The above-mentioned speech quality assessment method based on classification and feature extraction first classifies the speech data to be evaluated, then extracts features according to the classification results, and scores according to the extracted features. This fully considers the differences between speech signals and adopts different quality assessment methods for different speech signals. The evaluation effect is good and the algorithm complexity is low, and it has good practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flow chart of a method for evaluating speech quality based on classification and feature extraction in one embodiment;

[0055] Figure 2 A schematic diagram of an algorithm flow for performing actual operations using a voice quality assessment method in one embodiment;

[0056] Figure 3 is a structural block diagram of a speech quality assessment device based on classification and feature extraction in one embodiment;

[0057] Figure 4 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0059] like Figure 1 As shown, a speech quality assessment method based on classification and feature extraction is provided, comprising the following steps:

[0060] Step S100, obtaining speech data to be evaluated, and performing Fourier transform on each frame of speech in the speech data and then collecting corresponding statistics on its amplitude information to obtain amplitude information corresponding to each frame of speech;

[0061] Step S110, classifying each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include the first category, the second category and the third category, extracting features from each speech frame whose classification results are the second category and the third category, and evaluating the corresponding speech frame according to the extracted features to obtain the evaluation results of each speech frame of the second category and the third category;

[0062] Step S120: performing mean calculation based on the evaluation results of each speech frame of the second category and the third category to obtain the quality evaluation result of the speech data to be evaluated.

[0063] In this embodiment, the speech data to be evaluated is subjected to Fourier transform, and then the amplitude of each frame of speech data is calculated, and the speech data is classified according to the amplitude information of each frame, and then corresponding feature extraction is performed according to the frames of speech data classified into different categories. In this way, the differences between the frames of speech data are taken into account, and different methods are used for feature extraction and quality evaluation. Such a speech quality evaluation method has good evaluation effect and low algorithm complexity, and has good use value.

[0064] Specifically, in step S100, the acquired voice data y(n) is firstly converted into digital form, and the converted digital signal frame y is obtained. i (n), where i represents the frame number and n represents the number of sampling points. Then the digital signal is divided into frames y i (n) Perform Fourier transform to get Y i (k), where k represents the frequency point, and finally the amplitude information of each frame of speech data is counted, and in step S110, each frame of speech data is classified according to the amplitude information.

[0065] In step S110, each frame of speech is classified according to the amplitude information corresponding to each frame of speech and the preset parameters, specifically including: when the amplitude information corresponding to a frame of speech is less than the preset parameter, the frame of speech is classified into the first category. When the amplitude information corresponding to a frame of speech is greater than or equal to the preset parameter and less than 5 times the preset parameter, the frame of speech is classified into the second category. When the amplitude information corresponding to a frame of speech is greater than or equal to 5 times the preset parameter, the frame of speech is classified into the third category.

[0066] Specifically, when a frame of speech is smaller than a preset parameter, it means that there is no useful information in the frame data, so it is directly ignored and does not participate in the subsequent quality evaluation. The preset parameter is a very small positive number.

[0067] When the amplitude information corresponding to a frame of speech is greater than or equal to the preset parameter and less than 5 times the preset parameter, that is, when the frame of speech belongs to the second category, the frame of speech data is processed as follows: a frame of speech is filtered using a bandpass filter to obtain a filtered speech signal, feature extraction is performed based on the filtered speech signal to obtain a first feature that represents human auditory perception, the first feature is mapped to a quality space, and the evaluation result of the frame of speech is obtained.

[0068] Furthermore, before extracting features from the speech data of this category, the speech signal y is filtered through a bandpass filter L(h) with a very low cutoff frequency. i (n) is filtered so that the slowly changing or unchanged part of the spectrum is always suppressed, while the dynamic part is strengthened. The formula of the bandpass filter L(h) is as follows:

[0069]

[0070] The filtered speech signal is recorded as y2 i (n), and then extract a first feature F1 that can characterize human auditory perception based on the signal i , the feature extraction is expressed as:

[0071]

[0072] In formula (2), λ1 and λ2 are preset parameters, and their value range is between [0.1].

[0073] In this embodiment, through experimental analysis of a large number of voice videos, it is determined that the value range of λ1 and λ2 is between [0.1].

[0074] Then, the first feature is mapped to a quality space to obtain the evaluation result of the frame speech, using the following formula:

[0075]

[0076] In the above formula, MOS i is the evaluation result of each frame of speech.

[0077] When the amplitude information corresponding to a frame of speech is greater than or equal to 5 times the preset parameter, that is, when the frame of speech data belongs to the third category, the frame of speech data is processed as follows: a frame of speech is filtered using a GammaTone filter to obtain a filtered speech signal, and feature extraction is performed based on the filtered speech signal to obtain two second features and a third feature that represent human auditory perception, and an aggregation factor is calculated based on the second and third features, and then calculation is performed based on the aggregation factor to obtain an evaluation result of the frame of speech.

[0078] Further, the speech data classified into the third category is filtered through a GammaTone filter G(h) on the language signal y i (n) Filtering. The GammaTone filter only needs a few parameters to simulate the characteristics of human cochlear frequency decomposition. It is often used in audio signal processing to facilitate subsequent feature extraction. The formula of the filter G(h) is as follows:

[0079] G(h)=Ch m-1 exp(-2πbh)cos(2πf0+φ)D(h) (4)

[0080] In formula (4), C is the proportional coefficient, m is the filter order, b is the time attenuation coefficient, f0 is the filter center frequency, φ is the filter phase, when h>0, D(h)=1, otherwise D(h)=0, the filtered speech signal is recorded as y3 i (n).

[0081] The following two formulas are used to extract the second and third features of the filtered speech signal, where the second feature extraction is expressed as:

[0082] F2 i =λ3×exp(-1 / y3 i (n)) (5)

[0083] In formula (5), λ3 is a preset parameter whose value range is between [1, 10], and F2 i is the second feature;

[0084] The third feature extraction is expressed as:

[0085]

[0086] In formula (6), λ4 and λ5 are preset parameters, and their value range is between [10, 20]. i The third feature.

[0087] Specifically, the value ranges of parameters λ3, λ4 and λ5 are obtained through experimental analysis of a large number of voice videos.

[0088] According to the second feature F2 i And the third feature F3 i The following formula is used to calculate and obtain the aggregation factor μ1 corresponding to each frame of speech data: i :

[0089]

[0090] In formula (7), μ1 i Represents the aggregation factor.

[0091] Finally, the evaluation result of the frame speech is calculated based on the aggregation factor, using the following formula:

[0092]

[0093] In formula (8), MOS i is the evaluation result of each frame of speech.

[0094] In this way, each frame of classified speech data is calculated according to the quality calculation method of each category to obtain the corresponding quality evaluation MOS i , and then take the mean value to get the quality evaluation result of the speech signal to be evaluated.

[0095] In this embodiment, the duration of the voice data to be evaluated is less than 10 seconds.

[0096] When evaluating speech quality according to the above method, Figure 2 The algorithm flow shown is evaluated.

[0097] In the above-mentioned speech quality assessment method based on classification and feature extraction, the speech signal is first divided into three categories for classification and evaluation according to the frequency domain amplitude information: the first category does not participate in the evaluation; the second category signal is firstly subjected to speech enhancement through bandpass filtering, and then a feature that can characterize human auditory perception is extracted, and finally the quality is evaluated according to the feature; the third category signal is firstly processed through a GammaTone filter, and then two different features that can characterize human auditory perception are extracted, and the aggregation factor is calculated according to the two features, and finally the quality is evaluated according to the aggregation factor. The speech quality assessment method proposed in the present invention fully considers the differences between speech signals, adopts different quality assessment methods for different speech signals, has good evaluation effect and low algorithm complexity, and has good practical value.

[0098] It should be understood that although Figure 1-2The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1-2 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0099] In one embodiment, Figure 3 As shown, a speech quality assessment device based on classification and feature extraction is provided, comprising: a speech data acquisition module 200, a frame speech quality assessment module 210 and a quality assessment result acquisition module 220, wherein:

[0100] The speech data acquisition module 200 is used to acquire the speech data to be evaluated, and to perform Fourier transform on each frame of speech in the speech data and then to obtain the amplitude information corresponding to each frame of speech;

[0101] The speech quality assessment module 210 is used to classify each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include the first category, the second category and the third category, extract features of each speech frame whose classification results are the second category and the third category, and evaluate the corresponding speech frame according to the extracted features to obtain the evaluation results of each speech frame of the second category and the third category;

[0102] The quality assessment result obtaining module 220 is used to obtain the quality assessment result of the speech data to be assessed by performing mean calculation based on the assessment results of each speech frame of the second category and the third category.

[0103] The specific definition of the speech quality assessment device based on classification and feature extraction can be found in the above definition of the speech quality assessment method based on classification and feature extraction, which will not be repeated here. Each module in the above-mentioned speech quality assessment device based on classification and feature extraction can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0104] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a speech quality assessment method based on classification and feature extraction is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0105] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0106] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0107] Acquire speech data to be evaluated, and perform Fourier transform on each frame of speech in the speech data and then count the amplitude information thereof accordingly, so as to obtain amplitude information corresponding to each frame of speech;

[0108] Classifying each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include a first category, a second category and a third category, extracting features of each speech frame whose classification results are the second category and the third category, and evaluating the corresponding speech frame according to the extracted features to obtain evaluation results of each speech frame of the second category and the third category;

[0109] The quality assessment result of the speech data to be assessed is obtained by performing mean calculation based on the assessment results of each speech frame of the second category and the third category.

[0110] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0111] Acquire speech data to be evaluated, and perform Fourier transform on each frame of speech in the speech data and then count the amplitude information thereof accordingly, so as to obtain amplitude information corresponding to each frame of speech;

[0112] Classifying each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include a first category, a second category and a third category, extracting features of each speech frame whose classification results are the second category and the third category, and evaluating the corresponding speech frame according to the extracted features to obtain evaluation results of each speech frame of the second category and the third category;

[0113] The quality assessment result of the speech data to be assessed is obtained by performing mean calculation based on the assessment results of each speech frame of the second category and the third category.

[0114] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0115] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A speech quality assessment method based on classification and feature extraction, characterized in that: The method comprises: Acquire speech data to be evaluated, and perform Fourier transform on each frame of speech in the speech data and obtain corresponding statistical amplitude information to obtain amplitude information corresponding to each frame of speech; Classifying each speech frame according to the amplitude information corresponding to each speech frame and the preset parameters, the classification results include a first category, a second category and a third category, extracting features of each speech frame whose classification results are the second category and the third category, and evaluating the corresponding speech frame according to the extracted features to obtain evaluation results of each speech frame of the second category and the third category; The step of extracting features from each speech frame whose classification result is the second category and evaluating the corresponding speech frame according to the extracted features specifically includes: Filter each frame of speech using a bandpass filter to obtain a filtered speech signal; Performing feature extraction based on the filtered speech signal to obtain a first feature representing human auditory perception; Mapping the first features to a quality space respectively to obtain evaluation results of each frame of speech; The first feature extraction is expressed as: In the above formula, y2 i (n) is the filtered speech signal, λ1 and λ2 are preset parameters, and the value range of λ1 and λ2 is between [0, 1]. i is the first feature; The step of extracting features from each speech frame classified as the third category and evaluating the corresponding speech frame according to the extracted features specifically includes: Each frame of speech is filtered using a GammaTone filter to obtain a filtered speech signal; Performing feature extraction based on the filtered speech signal to obtain two second features and a third feature representing human auditory perception; The second feature extraction is expressed as: <h2 style=";text-align:left;direction:ltr">F2<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> =λ3×exp(-1 / y3<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (n)) In the above formula, y3 i (n) is the filtered speech signal, λ3 is a preset parameter whose value range is between [1, 10], F2 i is the second feature; The third feature extraction is expressed as: In the above formula, y3 i (n) is the filtered speech signal, λ4 and λ5 are preset parameters, whose value range is between [10, 20], and F3 i is the third feature; An aggregation factor is calculated based on the second feature and the third feature, and an evaluation result of each frame of speech is obtained by calculation based on the aggregation factor; and a quality evaluation result of the speech data to be evaluated is obtained by performing an average calculation based on the evaluation results of each frame of speech of the second category and the third category.

2. The method for evaluating speech quality according to claim 1, wherein: The classifying each frame of speech according to the amplitude information corresponding to each frame of speech and the preset parameters specifically includes: When the amplitude information corresponding to a frame of speech is less than the preset parameter, the frame of speech is classified into the first category; When the amplitude information corresponding to a frame of speech is greater than or equal to the preset parameter and less than 5 times the preset parameter, the frame of speech is classified into the second category; When the amplitude information corresponding to a frame of speech is greater than or equal to 5 times the preset parameter, the frame of speech is classified into the third category.

3. The method for evaluating speech quality according to claim 1, wherein: The first features are respectively mapped to a quality space to obtain the evaluation results of each frame of speech, using the following formula: In the above formula, MOS i is the evaluation result of each frame of speech.

4. The method for evaluating speech quality according to claim 3, characterized in that: The aggregation factor is calculated based on the second and third characteristics, wherein the following formula is used for calculation: In the above formula, μ1 i Represents the aggregation factor.

5. The method for evaluating speech quality according to claim 4, characterized in that: The evaluation result of each frame of speech is calculated according to the aggregation factor, using the following formula: In the above formula, MOS i is the evaluation result of each frame of speech.

6. The method for evaluating speech quality according to claim 1, wherein: The duration of the voice data to be evaluated is less than 10 seconds.

Images