Voice quality inspection method, device, equipment and storage medium based on artificial intelligence

Through the voice quality inspection method based on artificial intelligence, key audio is extracted and emotional and silent detection is performed, the problems of low accuracy and efficiency of voice quality inspection in the prior art are solved, and more accurate and efficient voice quality inspection results are achieved.

CN113903361BInactive Publication Date: 2024-11-12CHINA PING AN LIFE INSURANCE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111151525.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-29
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing voice quality inspection methods are inaccurate and efficient, making it difficult to extract key information from audio data, resulting in analyzing the entire audio, resulting in wasting resources and degrading accuracy.

Method used

Using artificial intelligence-based voice quality inspection method, key audio is extracted by comparing the audio duration with the preset time, and using pre-trained speech emotion recognition model and mute detection model to identify speech features and emotions, and eliminate user conversation information. Perform mute and speech speed detection to generate quality inspection results.

Benefits of technology

It improves the accuracy and efficiency of voice quality inspection, accurately identify target emotions and silent detection results, reduces the impact of dialogue information on detection results, enhances multi-dimensional analysis capabilities, and improves the accuracy of quality inspection results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113903361B_ABST
    Figure CN113903361B_ABST
Patent Text Reader

Abstract

The present invention relates to artificial intelligence, and provides a method, device, equipment and storage medium for speech quality inspection based on artificial intelligence. The method can extract key audio according to the audio duration and preset duration of the first user in multiple rounds of dialogue information, obtain a speech emotion recognition model, and the speech emotion recognition model includes a speech processing network and an emotion processing network, extract speech feature parameters in the key audio based on the speech processing network, recognize speech feature parameters and key audio based on the emotion processing network, obtain target emotion, remove the dialogue information of the first user in the key audio, obtain target audio, perform silence detection and speech speed detection on the target audio, obtain the silence detection result and speech speed detection result of the second user, generate quality inspection results according to the target emotion, silence detection result and speech speed detection result, and improve the accuracy and efficiency of speech quality inspection. In addition, the present invention also relates to blockchain technology, and the quality inspection results can be stored in the blockchain.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a speech quality inspection method, device, equipment and storage medium based on artificial intelligence. Background Art

[0002] In order to improve service quality and reduce service complaint rates, voice quality inspection of service conversation data is required. In current voice quality inspection solutions, a whole audio segment between a customer and customer service representative is usually directly analyzed according to preset rules to determine whether the customer service representative has responded to the customer's request according to the preset rules. Since the customer's request has a certain degree of flexibility, this quality inspection method has low accuracy. In addition, since it is difficult to extract key information from audio data, it is necessary to analyze a whole audio segment, resulting in low voice quality inspection efficiency. Summary of the invention

[0003] In view of the above, it is necessary to provide a voice quality inspection method, device, equipment and storage medium based on artificial intelligence, which can improve the accuracy and efficiency of voice quality inspection.

[0004] On the one hand, the present invention proposes a speech quality inspection method based on artificial intelligence, and the speech quality inspection method based on artificial intelligence includes:

[0005] When a quality inspection request is received, obtaining the audio to be inspected according to the quality inspection request, wherein the audio to be inspected includes information about multiple rounds of conversations between the first user and the second user;

[0006] Extracting key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and a preset duration;

[0007] Acquire a pre-trained speech emotion recognition model, wherein the speech emotion recognition model includes a speech processing network and an emotion processing network;

[0008] Extracting speech feature parameters in the key audio based on the speech processing network;

[0009] Based on the emotion processing network, the speech feature parameters and the key audio are identified to obtain the target emotion;

[0010] Eliminating the conversation information of the first user from the key audio to obtain the target audio;

[0011] Performing silence detection on the target audio to obtain a silence detection result of the second user;

[0012] Performing speech speed detection on the target audio to obtain a speech speed detection result of the second user;

[0013] A quality inspection result is generated according to the target emotion, the silence detection result and the speech speed detection result.

[0014] According to a preferred embodiment of the present invention, extracting key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and the preset duration includes:

[0015] Perform Fourier transform processing on any round of dialogue information to obtain multiple transformation results;

[0016] Mapping the multiple transformation results according to a preset feature template to obtain multiple audio features;

[0017] Calculating the feature similarity between each audio feature and a preset feature, and determining the audio feature with the greatest feature similarity as the user audio feature;

[0018] Counting the voice duration of the first user in each round of conversation information according to the user audio feature as the audio duration;

[0019] Determine the round in which the audio duration is greater than the preset duration as the target round;

[0020] The key audio is extracted from the audio to be inspected according to the target round.

[0021] According to a preferred embodiment of the present invention, extracting speech feature parameters in the key audio based on the speech processing network includes:

[0022] Obtaining a transfer function and a windowing function from the speech processing network;

[0023] Performing pre-emphasis processing on the key audio based on the transfer function to obtain a first audio;

[0024] Performing frame processing on the first audio to obtain multiple speech frames;

[0025] Obtaining a frame element of each speech frame, and calculating the product of the frame element and the windowing function to obtain a spectrogram;

[0026] The spectrogram is input into a convolutional feature extraction layer in the speech processing network to obtain the speech feature parameters.

[0027] According to a preferred embodiment of the present invention, the emotion processing network includes a tone recognition layer and a semantic parsing layer, and the identifying of the speech feature parameters and the key audio based on the emotion processing network to obtain the target emotion includes:

[0028] Recognize the speech feature parameters based on the tone recognition layer to obtain tone information;

[0029] Convert the key audio into text information, and parse the text information based on the semantic parsing layer to obtain semantic information;

[0030] Acquire a first emotion corresponding to the tone information, and acquire a second emotion corresponding to the semantic information;

[0031] If the first emotion is the same as the second emotion, determining the first emotion or the second emotion as the target emotion; or

[0032] If the first emotion is different from the second emotion, obtain the tone score corresponding to the tone information and the semantic score corresponding to the semantic information, weight and process the tone score and the semantic score to obtain the emotion score, and determine the type corresponding to the emotion score as the target emotion.

[0033] According to a preferred embodiment of the present invention, performing speech rate detection on the target audio to obtain a speech rate detection result of the second user includes:

[0034] Extracting information corresponding to the target audio from the text information as target text;

[0035] Counting the text volume of the target text, and counting the duration of the target audio as the target duration;

[0036] The ratio of the text amount to the target duration is calculated to obtain the speech rate detection result.

[0037] According to a preferred embodiment of the present invention, performing silence detection on the target audio to obtain a silence detection result of the second user includes:

[0038] Acquire multiple target frames from the target audio;

[0039] Extracting target speech features from each target frame;

[0040] Inputting the target speech feature into a pre-trained silence classifier to obtain a classification result for each target frame;

[0041] Detecting speech occurrence frames and speech disappearance frames in the target audio according to the classification result and the timing information of each target frame in the target audio;

[0042] Calculate the distance frame number between the speech disappearance frame and the speech appearance frame;

[0043] The ratio of the number of distance frames to the number of the plurality of target frames is calculated to obtain the silence detection result.

[0044] According to a preferred embodiment of the present invention, generating a quality inspection result according to the target emotion, the silence detection result and the speech speed detection result comprises:

[0045] Acquire a score interval corresponding to the target emotion, the score interval includes a first score and a second score, and the first score is less than the second score;

[0046] The first score, the silence detection result, and the speech speed detection result are calculated according to the following formula to obtain first quality inspection information:

[0047]

[0048] Wherein, y refers to the first quality inspection information, h refers to the first score, f refers to the silence detection result, and x refers to the speech speed detection result;

[0049] Analyze the second score, the silence detection result, and the speech speed detection result to obtain second quality inspection information;

[0050] The quality inspection result is generated according to the first quality inspection information and the second quality inspection information.

[0051] On the other hand, the present invention further proposes a speech quality inspection device based on artificial intelligence, the speech quality inspection device based on artificial intelligence comprising:

[0052] an acquisition unit, configured to acquire the audio to be inspected according to the quality inspection request when a quality inspection request is received, wherein the audio to be inspected includes information about multiple rounds of conversations between the first user and the second user;

[0053] An extraction unit, configured to extract key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and a preset duration;

[0054] The acquisition unit is further used to acquire a pre-trained speech emotion recognition model, wherein the speech emotion recognition model includes a speech processing network and an emotion processing network;

[0055] The extraction unit is further used to extract speech feature parameters in the key audio based on the speech processing network;

[0056] A recognition unit, configured to recognize the speech feature parameters and the key audio based on the emotion processing network to obtain a target emotion;

[0057] A removing unit, used for removing the conversation information of the first user from the key audio to obtain the target audio;

[0058] a detection unit, configured to perform silence detection on the target audio to obtain a silence detection result of the second user;

[0059] The detection unit is further configured to perform speech speed detection on the target audio to obtain a speech speed detection result of the second user;

[0060] A generating unit is used to generate a quality inspection result according to the target emotion, the silence detection result and the speech speed detection result.

[0061] On the other hand, the present invention further provides an electronic device, comprising:

[0062] a memory storing computer readable instructions; and

[0063] A processor executes the computer-readable instructions stored in the memory to implement the artificial intelligence-based speech quality inspection method.

[0064] On the other hand, the present invention also proposes a computer-readable storage medium, in which computer-readable instructions are stored. The computer-readable instructions are executed by a processor in an electronic device to implement the artificial intelligence-based speech quality inspection method.

[0065] It can be seen from the above technical scheme that the present invention extracts the key audio by comparing the audio duration with the preset duration, so as to extract the key information in the audio to be inspected, avoid analyzing the entire audio in the audio to be inspected, and improve the quality inspection efficiency. Furthermore, based on the speech processing network, the speech feature parameters can be accurately extracted to improve the recognition accuracy of the speech feature parameters by the emotion processing network, so as to accurately identify the target emotion corresponding to the target audio. Furthermore, by performing silence detection and speech speed detection on the target audio, the influence of the conversation information of the first user on the silence detection result and the speech speed detection result is eliminated, so as to improve the accuracy of the silence detection result and the speech speed detection result. The quality inspection result of the audio to be inspected can be analyzed from multiple dimensions through the target emotion, the silence detection result and the speech speed detection result, thereby improving the accuracy of the quality inspection result. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a flow chart of a preferred embodiment of the speech quality inspection method based on artificial intelligence of the present invention.

[0067] Figure 2 It is a functional module diagram of a preferred embodiment of the speech quality inspection device based on artificial intelligence of the present invention.

[0068] Figure 3 It is a structural schematic diagram of an electronic device of a preferred embodiment of the present invention to implement the voice quality inspection method based on artificial intelligence. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0070] like Figure 1 FIG. 1 is a flow chart of a preferred embodiment of the speech quality inspection method based on artificial intelligence of the present invention. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted.

[0071] The speech quality inspection method based on artificial intelligence can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0072] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0073] The artificial intelligence-based speech quality inspection method is applied to one or more electronic devices, which are devices that can automatically perform numerical calculations and / or information processing according to pre-set or stored computer-readable instructions, and whose hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0074] The electronic device may be any electronic product that can perform human-computer interaction with a user, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), a game console, an interactive network television (IPTV), a smart wearable device, etc.

[0075] The electronic device may include a network device and / or a user device. The network device includes, but is not limited to, a single network electronic device, an electronic device group consisting of multiple network electronic devices, or a cloud consisting of a large number of hosts or network electronic devices based on cloud computing.

[0076] The network where the electronic device is located includes, but is not limited to: the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VPN), etc.

[0077] S10, when a quality inspection request is received, obtaining audio to be inspected according to the quality inspection request, wherein the audio to be inspected includes information of multiple rounds of conversations between the first user and the second user.

[0078] In at least one embodiment of the present invention, the quality inspection request may be triggered and generated by a supervisory user who is responsible for customer service quality, and the quality inspection request may also be triggered and generated at a preset time point.

[0079] The quality inspection request carries an audio path.

[0080] The audio to be quality-checked refers to an audio segment that needs to be tested for service quality.

[0081] The first user may be a customer, and the second user may be a customer service representative.

[0082] In at least one embodiment of the present invention, the electronic device obtaining the audio to be quality checked according to the quality check request includes:

[0083] Parsing the message of the quality inspection request to obtain the data information;

[0084] Extracting path information from the data information;

[0085] The audio to be quality inspected is randomly acquired according to the path information.

[0086] The path information stores a plurality of audios that have not been quality-checked.

[0087] The audio to be inspected can be quickly acquired through the path information.

[0088] S11, extracting key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and a preset duration.

[0089] In at least one embodiment of the present invention, the audio duration refers to the voice duration of the first user in each round of dialogue information.

[0090] The preset duration can be determined based on the average duration when users make requests in specific business scenarios.

[0091] The key audio includes key information of the communication process between the first user and the second user.

[0092] In at least one embodiment of the present invention, the electronic device extracting key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and the preset duration includes:

[0093] Perform Fourier transform processing on any round of dialogue information to obtain multiple transformation results;

[0094] Mapping the multiple transformation results according to a preset feature template to obtain multiple audio features;

[0095] Calculating the feature similarity between each audio feature and a preset feature, and determining the audio feature with the greatest feature similarity as the user audio feature;

[0096] Counting the voice duration of the first user in each round of conversation information according to the user audio feature as the audio duration;

[0097] Determine the round in which the audio duration is greater than the preset duration as the target round;

[0098] The key audio is extracted from the audio to be inspected according to the target round.

[0099] Among them, the multiple transformation results refer to the information obtained after the Fourier transform processing of multiple audio sentences in any round of dialogue information. The transformation results corresponding to the same text sentence by different users are different, and the transformation results corresponding to different text sentences by the same user are also different. The transformation results corresponding to the sentence audios emitted by the same user for different text sentences belong to the same numerical interval in the preset feature template. For example, the transformation result corresponding to the text sentence "We have no way to do this" by user A is 0.31, the transformation result corresponding to the text sentence "This can be handled like this" by user A is 0.35, and the transformation result corresponding to the text sentence "We have no way to do this" by user B is 0.41. The preset feature template includes: the feature corresponding to the transformation numerical interval (0.30, 0.35] is feature abc.

[0100] The plurality of audio features include timbre information of different users.

[0101] By performing Fourier transform processing on any round of dialogue information and mapping the multiple transformation results based on the preset feature template, the influence of the difference in sentence audio uttered by the same user for different text sentences can be eliminated, thereby improving the accuracy of extracting the multiple audio features, and further, by comparing the audio duration with the preset duration, the key audio can be quickly extracted from the audio to be inspected.

[0102] S12, obtaining a pre-trained speech emotion recognition model, wherein the speech emotion recognition model includes a speech processing network and an emotion processing network.

[0103] In at least one embodiment of the present invention, the speech emotion recognition model can be used to recognize emotion information corresponding to the audio.

[0104] The speech processing network is used to extract feature parameters in audio.

[0105] The emotion processing network is used to analyze the audio emotion from the dimension of the characteristic parameters, and to analyze the audio emotion from the dimension of the audio.

[0106] S13, extracting speech feature parameters in the key audio based on the speech processing network.

[0107] In at least one embodiment of the present invention, the speech feature parameter refers to parameter information that can be used to identify tone information.

[0108] In at least one embodiment of the present invention, the electronic device extracting the speech feature parameters in the key audio based on the speech processing network includes:

[0109] Obtaining a transfer function and a windowing function from the speech processing network;

[0110] Performing pre-emphasis processing on the key audio based on the transfer function to obtain a first audio;

[0111] Performing frame processing on the first audio to obtain multiple speech frames;

[0112] Obtaining a frame element of each speech frame, and calculating the product of the frame element and the windowing function to obtain a spectrogram;

[0113] The spectrogram is input into a convolutional feature extraction layer in the speech processing network to obtain the speech feature parameters.

[0114] The transfer function varies with the network parameters in the speech processing network. For example, if the pre-emphasis coefficient in the speech processing network 1 is 0.92, then the transfer function is The pre-emphasis coefficient in the speech processing network 2 is 0.98, so the transfer function is

[0115] The convolutional feature extraction layer includes multiple convolutional neural networks.

[0116] By emphasizing the key audio through the transfer function, the high-frequency part of the key audio can be emphasized, thereby improving the high-frequency resolution of the key audio, thereby improving the extraction accuracy of the speech feature parameters. By performing frame processing on the first audio, the continuity of information in each speech frame can be ensured, and the extraction efficiency of the speech feature parameters can be improved. By performing windowing processing on each speech frame, the sampled speech waveform can be emphasized, further improving the extraction accuracy of the speech feature parameters.

[0117] S14, identifying the speech feature parameters and the key audio based on the emotion processing network to obtain the target emotion.

[0118] In at least one embodiment of the present invention, the target emotion includes, but is not limited to: anger, happiness, fear, sadness, surprise, neutrality, etc.

[0119] In at least one embodiment of the present invention, the emotion processing network includes a tone recognition layer and a semantic parsing layer, and the electronic device recognizes the speech feature parameters and the key audio based on the emotion processing network, and obtains the target emotion including:

[0120] Recognize the speech feature parameters based on the tone recognition layer to obtain tone information;

[0121] Convert the key audio into text information, and parse the text information based on the semantic parsing layer to obtain semantic information;

[0122] Acquire a first emotion corresponding to the tone information, and acquire a second emotion corresponding to the semantic information;

[0123] If the first emotion is the same as the second emotion, determining the first emotion or the second emotion as the target emotion; or

[0124] If the first emotion is different from the second emotion, obtain the tone score corresponding to the tone information and the semantic score corresponding to the semantic information, weight and process the tone score and the semantic score to obtain the emotion score, and determine the type corresponding to the emotion score as the target emotion.

[0125] The tone information is used to characterize the tone in the key audio.

[0126] The semantic information is used to represent the semantics of the text information.

[0127] By identifying the voice feature parameters through the tone recognition layer and parsing the text information through the semantic parsing layer, the emotion of the key audio can be analyzed from multiple dimensions to improve the accuracy of the target emotion.

[0128] S15, removing the conversation information of the first user from the key audio to obtain the target audio.

[0129] In at least one embodiment of the present invention, the target audio refers to audio that does not contain conversation information of the first user.

[0130] In at least one embodiment of the present invention, the electronic device removes the conversation information of the first user from the key audio to obtain the target audio, comprising:

[0131] Detecting the key audio according to the user audio feature to obtain the conversation information of the first user;

[0132] The conversation information of the first user is removed from the key audio to obtain the target audio.

[0133] The conversation information of the first user can be accurately detected through the user audio features, thereby improving the accuracy of the target audio.

[0134] S16, performing silence detection on the target audio to obtain a silence detection result of the second user.

[0135] In at least one embodiment of the present invention, the silence detection result refers to the proportion of silence frames in the target audio.

[0136] In at least one embodiment of the present invention, the electronic device performs silence detection on the target audio, and obtaining the silence detection result of the second user includes:

[0137] Acquire multiple target frames from the target audio;

[0138] Extracting target speech features from each target frame;

[0139] Inputting the target speech feature into a pre-trained silence classifier to obtain a classification result for each target frame;

[0140] Detecting speech occurrence frames and speech disappearance frames in the target audio according to the classification result and the timing information of each target frame in the target audio;

[0141] Calculate the distance frame number between the speech disappearance frame and the speech appearance frame;

[0142] The ratio of the number of distance frames to the number of the plurality of target frames is calculated to obtain the silence detection result.

[0143] The silence classifier may be generated based on a VAD algorithm.

[0144] The category results include whether the target frame is a silence frame and whether the target frame is not a silence frame.

[0145] The timing information refers to the position sequence number of the target frame in the target audio.

[0146] The speech occurrence frame refers to a target frame for conversion from a mute result to a non-mute result, and the speech disappearance frame refers to a target frame for conversion from a non-mute result to a mute result.

[0147] For example, there are 5 target frames in the target audio, and the category results of the 5 target frames are: the first target frame is a silent frame, the second target frame is a silent frame, the third target frame is not a silent frame, the fourth target frame is not a silent frame, and the fifth target frame is not a silent frame. Then the voice disappearance frame includes: the first frame, and the voice appearance frame includes: the third frame. Then the distance frame number is 2, and the silence detection result is that the proportion of silent frames in the target audio is 0.4.

[0148] By analyzing the multiple target frames frame by frame by the silence classifier, the accuracy of the category result can be improved, and then the voice appearance frame and the voice disappearance frame can be accurately generated according to the category result and the timing information of each target frame in the target audio, which is conducive to the rapid generation of the silence detection result.

[0149] S17, performing speech speed detection on the target audio to obtain a speech speed detection result of the second user.

[0150] In at least one embodiment of the present invention, the speech rate detection result is used to indicate the pronunciation speed of the second user.

[0151] In at least one embodiment of the present invention, the electronic device performs speech speed detection on the target audio, and obtaining the speech speed detection result of the second user includes:

[0152] Extracting information corresponding to the target audio from the text information as target text;

[0153] Counting the text volume of the target text, and counting the duration of the target audio as the target duration;

[0154] The ratio of the text amount to the target duration is calculated to obtain the speech rate detection result.

[0155] By extracting the target text from the text information, there is no need to convert the target audio into text again, so the efficiency of extracting the target text can be improved, thereby improving the efficiency of generating the speech rate detection result.

[0156] S18, generating a quality inspection result according to the target emotion, the silence detection result and the speech speed detection result.

[0157] In at least one embodiment of the present invention, the quality inspection result refers to the quality inspection result of the audio to be inspected. The quality inspection result may be a numerical range. For example, the quality inspection result is [0.82, 0.86].

[0158] It should be emphasized that in order to further ensure the privacy and security of the above quality inspection results, the above quality inspection results can also be stored in a blockchain node.

[0159] In at least one embodiment of the present invention, the electronic device generating a quality inspection result according to the target emotion, the silence detection result, and the speech speed detection result comprises:

[0160] Acquire a score interval corresponding to the target emotion, the score interval includes a first score and a second score, and the first score is less than the second score;

[0161] The first score, the silence detection result, and the speech speed detection result are calculated according to the following formula to obtain first quality inspection information:

[0162]

[0163] Wherein, y refers to the first quality inspection information, h refers to the first score, f refers to the silence detection result, and x refers to the speech speed detection result;

[0164] Analyze the second score, the silence detection result, and the speech speed detection result to obtain second quality inspection information;

[0165] The quality inspection result is generated according to the first quality inspection information and the second quality inspection information.

[0166] Through the above implementation, the audio to be quality inspected can be analyzed according to information of different dimensions, which can improve the accuracy of the quality inspection result.

[0167] It can be seen from the above technical scheme that the present invention extracts the key audio by comparing the audio duration with the preset duration, so as to extract the key information in the audio to be inspected, avoid analyzing the entire audio in the audio to be inspected, and improve the quality inspection efficiency. Furthermore, based on the speech processing network, the speech feature parameters can be accurately extracted to improve the recognition accuracy of the speech feature parameters by the emotion processing network, so as to accurately identify the target emotion corresponding to the target audio. Furthermore, by performing silence detection and speech speed detection on the target audio, the influence of the conversation information of the first user on the silence detection result and the speech speed detection result is eliminated, so as to improve the accuracy of the silence detection result and the speech speed detection result. The quality inspection result of the audio to be inspected can be analyzed from multiple dimensions through the target emotion, the silence detection result and the speech speed detection result, thereby improving the accuracy of the quality inspection result.

[0168] like Figure 2 As shown, it is a functional module diagram of a preferred embodiment of the speech quality inspection device based on artificial intelligence of the present invention. The speech quality inspection device based on artificial intelligence 11 includes an acquisition unit 110, an extraction unit 111, a recognition unit 112, a rejection unit 113, a detection unit 114 and a generation unit 115. The module / unit referred to in the present invention refers to a series of computer-readable instruction segments that can be acquired by the processor 13 and can complete fixed functions, which are stored in the memory 12. In this embodiment, the functions of each module / unit will be described in detail in subsequent embodiments.

[0169] When a quality inspection request is received, the acquisition unit 110 acquires the audio to be inspected according to the quality inspection request, where the audio to be inspected includes multiple rounds of conversation information between the first user and the second user.

[0170] In at least one embodiment of the present invention, the quality inspection request may be triggered and generated by a supervisory user who is responsible for customer service quality, and the quality inspection request may also be triggered and generated at a preset time point.

[0171] The quality inspection request carries an audio path.

[0172] The audio to be quality-checked refers to an audio segment that needs to be tested for service quality.

[0173] The first user may be a customer, and the second user may be a customer service representative.

[0174] In at least one embodiment of the present invention, the acquiring unit 110 acquires the audio to be quality checked according to the quality check request, including:

[0175] Parsing the message of the quality inspection request to obtain the data information;

[0176] Extracting path information from the data information;

[0177] The audio to be quality inspected is randomly acquired according to the path information.

[0178] The path information stores a plurality of audios that have not been quality-checked.

[0179] The audio to be inspected can be quickly acquired through the path information.

[0180] The extraction unit 111 extracts key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and a preset duration.

[0181] In at least one embodiment of the present invention, the audio duration refers to the voice duration of the first user in each round of dialogue information.

[0182] The preset duration can be determined based on the average duration when users make requests in specific business scenarios.

[0183] The key audio includes key information of the communication process between the first user and the second user.

[0184] In at least one embodiment of the present invention, the extracting unit 111 extracts key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and the preset duration, including:

[0185] Perform Fourier transform processing on any round of dialogue information to obtain multiple transformation results;

[0186] Mapping the multiple transformation results according to a preset feature template to obtain multiple audio features;

[0187] Calculating the feature similarity between each audio feature and a preset feature, and determining the audio feature with the greatest feature similarity as the user audio feature;

[0188] Counting the voice duration of the first user in each round of conversation information according to the user audio feature as the audio duration;

[0189] Determine the round in which the audio duration is greater than the preset duration as the target round;

[0190] The key audio is extracted from the audio to be inspected according to the target round.

[0191] Among them, the multiple transformation results refer to the information obtained after the Fourier transform processing of multiple audio sentences in any round of dialogue information. The transformation results corresponding to the same text sentence by different users are different, and the transformation results corresponding to different text sentences by the same user are also different. The transformation results corresponding to the sentence audios emitted by the same user for different text sentences belong to the same numerical interval in the preset feature template. For example, the transformation result corresponding to the text sentence "We have no way to do this" by user A is 0.31, the transformation result corresponding to the text sentence "This can be handled like this" by user A is 0.35, and the transformation result corresponding to the text sentence "We have no way to do this" by user B is 0.41. The preset feature template includes: the feature corresponding to the transformation numerical interval (0.30, 0.35] is feature abc.

[0192] The plurality of audio features include timbre information of different users.

[0193] By performing Fourier transform processing on any round of dialogue information and mapping the multiple transformation results based on the preset feature template, the influence of the difference in sentence audio uttered by the same user for different text sentences can be eliminated, thereby improving the accuracy of extracting the multiple audio features, and further, by comparing the audio duration with the preset duration, the key audio can be quickly extracted from the audio to be inspected.

[0194] The acquisition unit 110 acquires a pre-trained speech emotion recognition model, where the speech emotion recognition model includes a speech processing network and an emotion processing network.

[0195] In at least one embodiment of the present invention, the speech emotion recognition model can be used to recognize emotion information corresponding to the audio.

[0196] The speech processing network is used to extract feature parameters in audio.

[0197] The emotion processing network is used to analyze the audio emotion from the dimension of the characteristic parameters, and to analyze the audio emotion from the dimension of the audio.

[0198] The extraction unit 111 extracts speech feature parameters in the key audio based on the speech processing network.

[0199] In at least one embodiment of the present invention, the speech feature parameter refers to parameter information that can be used to identify tone information.

[0200] In at least one embodiment of the present invention, the extraction unit 111 extracts the speech feature parameters in the key audio based on the speech processing network, including:

[0201] Obtaining a transfer function and a windowing function from the speech processing network;

[0202] Performing pre-emphasis processing on the key audio based on the transfer function to obtain a first audio;

[0203] Performing frame processing on the first audio to obtain multiple speech frames;

[0204] Obtaining a frame element of each speech frame, and calculating the product of the frame element and the windowing function to obtain a spectrogram;

[0205] The spectrogram is input into a convolutional feature extraction layer in the speech processing network to obtain the speech feature parameters.

[0206] The transfer function varies with the network parameters in the speech processing network. For example, if the pre-emphasis coefficient in the speech processing network 1 is 0.92, then the transfer function is The pre-emphasis coefficient in the speech processing network 2 is 0.98, so the transfer function is

[0207] The convolutional feature extraction layer includes multiple convolutional neural networks.

[0208] By emphasizing the key audio through the transfer function, the high-frequency part of the key audio can be emphasized, thereby improving the high-frequency resolution of the key audio, thereby improving the extraction accuracy of the speech feature parameters. By performing frame processing on the first audio, the continuity of information in each speech frame can be ensured, and the extraction efficiency of the speech feature parameters can be improved. By performing windowing processing on each speech frame, the sampled speech waveform can be emphasized, further improving the extraction accuracy of the speech feature parameters.

[0209] The recognition unit 112 recognizes the speech feature parameters and the key audio based on the emotion processing network to obtain the target emotion.

[0210] In at least one embodiment of the present invention, the target emotion includes, but is not limited to: anger, happiness, fear, sadness, surprise, neutrality, etc.

[0211] In at least one embodiment of the present invention, the emotion processing network includes a tone recognition layer and a semantic parsing layer, and the recognition unit 112 recognizes the speech feature parameters and the key audio based on the emotion processing network, and obtains the target emotion including:

[0212] Recognize the speech feature parameters based on the tone recognition layer to obtain tone information;

[0213] Convert the key audio into text information, and parse the text information based on the semantic parsing layer to obtain semantic information;

[0214] Acquire a first emotion corresponding to the tone information, and acquire a second emotion corresponding to the semantic information;

[0215] If the first emotion is the same as the second emotion, determining the first emotion or the second emotion as the target emotion; or

[0216] If the first emotion is different from the second emotion, obtain the tone score corresponding to the tone information and the semantic score corresponding to the semantic information, weight and process the tone score and the semantic score to obtain the emotion score, and determine the type corresponding to the emotion score as the target emotion.

[0217] The tone information is used to characterize the tone in the key audio.

[0218] The semantic information is used to represent the semantics of the text information.

[0219] By identifying the voice feature parameters through the tone recognition layer and parsing the text information through the semantic parsing layer, the emotion of the key audio can be analyzed from multiple dimensions to improve the accuracy of the target emotion.

[0220] The elimination unit 113 eliminates the conversation information of the first user in the key audio to obtain the target audio.

[0221] In at least one embodiment of the present invention, the target audio refers to audio that does not contain conversation information of the first user.

[0222] In at least one embodiment of the present invention, the removing unit 113 removes the conversation information of the first user from the key audio to obtain the target audio, including:

[0223] Detecting the key audio according to the user audio feature to obtain the conversation information of the first user;

[0224] The conversation information of the first user is removed from the key audio to obtain the target audio.

[0225] The conversation information of the first user can be accurately detected through the user audio features, thereby improving the accuracy of the target audio.

[0226] The detection unit 114 performs silence detection on the target audio to obtain a silence detection result of the second user.

[0227] In at least one embodiment of the present invention, the silence detection result refers to the proportion of silence frames in the target audio.

[0228] In at least one embodiment of the present invention, the detection unit 114 performs silence detection on the target audio, and obtaining the silence detection result of the second user includes:

[0229] Acquire multiple target frames from the target audio;

[0230] Extracting target speech features from each target frame;

[0231] Inputting the target speech feature into a pre-trained silence classifier to obtain a classification result for each target frame;

[0232] Detecting speech occurrence frames and speech disappearance frames in the target audio according to the classification result and the timing information of each target frame in the target audio;

[0233] Calculate the distance frame number between the speech disappearance frame and the speech appearance frame;

[0234] The ratio of the number of distance frames to the number of the plurality of target frames is calculated to obtain the silence detection result.

[0235] The silence classifier may be generated based on a VAD algorithm.

[0236] The category results include whether the target frame is a silence frame and whether the target frame is not a silence frame.

[0237] The timing information refers to the position sequence number of the target frame in the target audio.

[0238] The speech occurrence frame refers to a target frame for conversion from a mute result to a non-mute result, and the speech disappearance frame refers to a target frame for conversion from a non-mute result to a mute result.

[0239] For example, there are 5 target frames in the target audio, and the category results of the 5 target frames are: the first target frame is a silent frame, the second target frame is a silent frame, the third target frame is not a silent frame, the fourth target frame is not a silent frame, and the fifth target frame is not a silent frame. Then the voice disappearance frame includes: the first frame, and the voice appearance frame includes: the third frame. Then the distance frame number is 2, and the silence detection result is that the proportion of silent frames in the target audio is 0.4.

[0240] By analyzing the multiple target frames frame by frame by the silence classifier, the accuracy of the category result can be improved, and then the voice appearance frame and the voice disappearance frame can be accurately generated according to the category result and the timing information of each target frame in the target audio, which is conducive to the rapid generation of the silence detection result.

[0241] The detection unit 114 performs speech rate detection on the target audio to obtain a speech rate detection result of the second user.

[0242] In at least one embodiment of the present invention, the speech rate detection result is used to indicate the pronunciation speed of the second user.

[0243] In at least one embodiment of the present invention, the detection unit 114 performs speech rate detection on the target audio, and obtaining the speech rate detection result of the second user includes:

[0244] Extracting information corresponding to the target audio from the text information as target text;

[0245] Counting the text volume of the target text, and counting the duration of the target audio as the target duration;

[0246] The ratio of the text amount to the target duration is calculated to obtain the speech rate detection result.

[0247] By extracting the target text from the text information, there is no need to convert the target audio into text again, so the efficiency of extracting the target text can be improved, thereby improving the efficiency of generating the speech rate detection result.

[0248] The generating unit 115 generates a quality inspection result according to the target emotion, the silence detection result and the speech rate detection result.

[0249] In at least one embodiment of the present invention, the quality inspection result refers to the quality inspection result of the audio to be inspected. The quality inspection result may be a numerical range. For example, the quality inspection result is [0.82, 0.86].

[0250] It should be emphasized that in order to further ensure the privacy and security of the above quality inspection results, the above quality inspection results can also be stored in a blockchain node.

[0251] In at least one embodiment of the present invention, the generating unit 115 generates the quality inspection result according to the target emotion, the silence detection result and the speech rate detection result, including:

[0252] Acquire a score interval corresponding to the target emotion, the score interval includes a first score and a second score, and the first score is less than the second score;

[0253] The first score, the silence detection result, and the speech speed detection result are calculated according to the following formula to obtain first quality inspection information:

[0254]

[0255] Wherein, y refers to the first quality inspection information, h refers to the first score, f refers to the silence detection result, and x refers to the speech speed detection result;

[0256] Analyze the second score, the silence detection result, and the speech speed detection result to obtain second quality inspection information;

[0257] The quality inspection result is generated according to the first quality inspection information and the second quality inspection information.

[0258] Through the above implementation, the audio to be quality inspected can be analyzed according to information of different dimensions, which can improve the accuracy of the quality inspection result.

[0259] It can be seen from the above technical scheme that the present invention extracts the key audio by comparing the audio duration with the preset duration, so as to extract the key information in the audio to be inspected, avoid analyzing the entire audio in the audio to be inspected, and improve the quality inspection efficiency. Furthermore, based on the speech processing network, the speech feature parameters can be accurately extracted to improve the recognition accuracy of the speech feature parameters by the emotion processing network, so as to accurately identify the target emotion corresponding to the target audio. Furthermore, by performing silence detection and speech speed detection on the target audio, the influence of the conversation information of the first user on the silence detection result and the speech speed detection result is eliminated, so as to improve the accuracy of the silence detection result and the speech speed detection result. The quality inspection result of the audio to be inspected can be analyzed from multiple dimensions through the target emotion, the silence detection result and the speech speed detection result, thereby improving the accuracy of the quality inspection result.

[0260] like Figure 3 , which is a schematic diagram of the structure of an electronic device of a preferred embodiment of the present invention for implementing a speech quality inspection method based on artificial intelligence.

[0261] In one embodiment of the present invention, the electronic device 1 includes, but is not limited to, a memory 12, a processor 13, and computer-readable instructions stored in the memory 12 and executable on the processor 13, such as an artificial intelligence-based voice quality inspection program.

[0262] Those skilled in the art will understand that the schematic diagram is merely an example of the electronic device 1 and does not constitute a limitation on the electronic device 1. The electronic device 1 may include more or fewer components than shown in the diagram, or a combination of certain components, or different components. For example, the electronic device 1 may also include input and output devices, network access devices, buses, etc.

[0263] The processor 13 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor 13 is the computing core and control center of the electronic device 1, and uses various interfaces and lines to connect various parts of the entire electronic device 1, and executes the operating system of the electronic device 1 and various installed applications, program codes, etc.

[0264] Exemplarily, the computer-readable instructions may be divided into one or more modules / units, which are stored in the memory 12 and executed by the processor 13 to complete the present invention. The one or more modules / units may be a series of computer-readable instruction segments capable of completing specific functions, which are used to describe the execution process of the computer-readable instructions in the electronic device 1. For example, the computer-readable instructions may be divided into an acquisition unit 110, an extraction unit 111, a recognition unit 112, a rejection unit 113, a detection unit 114, and a generation unit 115.

[0265] The memory 12 can be used to store the computer-readable instructions and / or modules, and the processor 13 implements various functions of the electronic device 1 by running or executing the computer-readable instructions and / or modules stored in the memory 12, and calling the data stored in the memory 12. The memory 12 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the electronic device, etc. The memory 12 can include non-volatile and volatile memories, such as: a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other storage devices.

[0266] The memory 12 may be an external memory and / or an internal memory of the electronic device 1. Furthermore, the memory 12 may be a memory in a physical form, such as a memory stick, a TF card (Trans-flash Card), and the like.

[0267] If the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also instruct the relevant hardware to complete it through computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium. When the computer-readable instructions are executed by the processor, the steps of the above-mentioned various method embodiments can be implemented.

[0268] The computer-readable instructions include computer-readable instruction codes, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer-readable instruction code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM).

[0269] The blockchain referred to in this invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0270] Combination Figure 1 The memory 12 in the electronic device 1 stores computer-readable instructions to implement a voice quality inspection method based on artificial intelligence, and the processor 13 can execute the computer-readable instructions to implement:

[0271] When a quality inspection request is received, obtaining the audio to be inspected according to the quality inspection request, wherein the audio to be inspected includes information about multiple rounds of conversations between the first user and the second user;

[0272] Extracting key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and a preset duration;

[0273] Acquire a pre-trained speech emotion recognition model, wherein the speech emotion recognition model includes a speech processing network and an emotion processing network;

[0274] Extracting speech feature parameters in the key audio based on the speech processing network;

[0275] Based on the emotion processing network, the speech feature parameters and the key audio are identified to obtain the target emotion;

[0276] Eliminating the conversation information of the first user from the key audio to obtain the target audio;

[0277] Performing silence detection on the target audio to obtain a silence detection result of the second user;

[0278] Performing speech speed detection on the target audio to obtain a speech speed detection result of the second user;

[0279] A quality inspection result is generated according to the target emotion, the silence detection result and the speech speed detection result.

[0280] Specifically, the specific implementation method of the processor 13 for the above-mentioned computer readable instructions can refer to Figure 1 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0281] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0282] The computer-readable storage medium stores computer-readable instructions, wherein the computer-readable instructions are used to implement the following steps when executed by the processor 13:

[0283] When a quality inspection request is received, obtaining the audio to be inspected according to the quality inspection request, wherein the audio to be inspected includes information about multiple rounds of conversations between the first user and the second user;

[0284] Extracting key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and a preset duration;

[0285] Acquire a pre-trained speech emotion recognition model, wherein the speech emotion recognition model includes a speech processing network and an emotion processing network;

[0286] Extracting speech feature parameters in the key audio based on the speech processing network;

[0287] Based on the emotion processing network, the speech feature parameters and the key audio are identified to obtain the target emotion;

[0288] Eliminating the conversation information of the first user from the key audio to obtain the target audio;

[0289] Performing silence detection on the target audio to obtain a silence detection result of the second user;

[0290] Performing speech speed detection on the target audio to obtain a speech speed detection result of the second user;

[0291] A quality inspection result is generated according to the target emotion, the silence detection result and the speech speed detection result.

[0292] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0293] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0294] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0295] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices mentioned can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0296] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A speech quality inspection method based on artificial intelligence, characterized in that: The speech quality inspection method based on artificial intelligence includes: When a quality inspection request is received, obtaining the audio to be inspected according to the quality inspection request, wherein the audio to be inspected includes information about multiple rounds of conversations between the first user and the second user; Extracting key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and a preset duration; Acquire a pre-trained speech emotion recognition model, wherein the speech emotion recognition model includes a speech processing network and an emotion processing network; Extracting speech feature parameters in the key audio based on the speech processing network; Based on the emotion processing network, the speech feature parameters and the key audio are identified to obtain the target emotion; Eliminating the conversation information of the first user from the key audio to obtain the target audio; Performing silence detection on the target audio to obtain a silence detection result of the second user includes: obtaining multiple target frames from the target audio; extracting target speech features from each target frame; inputting the target speech features into a pre-trained silence classifier to obtain a category result for each target frame; detecting speech occurrence frames and speech disappearance frames in the target audio according to the category result and the timing information of each target frame in the target audio; calculating the number of distance frames between the speech disappearance frame and the speech occurrence frame; calculating the ratio of the number of distance frames to the number of the multiple target frames to obtain the silence detection result; Performing speech speed detection on the target audio to obtain a speech speed detection result of the second user; Generating a quality inspection result according to the target emotion, the silence detection result, and the speech speed detection result, including: obtaining a score interval corresponding to the target emotion, the score interval including a first score and a second score, the first score being less than the second score; calculating the first score, the silence detection result, and the speech speed detection result according to the following formula to obtain first quality inspection information: Among them, y refers to the first quality inspection information, h refers to the first score, f refers to the silence detection result, and x refers to the speech rate detection result; the second score, the silence detection result and the speech rate detection result are analyzed to obtain the second quality inspection information; the quality inspection result is generated according to the first quality inspection information and the second quality inspection information.

2. The method for speech quality inspection based on artificial intelligence according to claim 1, characterized in that: The extracting key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and the preset duration includes: Perform Fourier transform processing on any round of dialogue information to obtain multiple transformation results; Mapping the multiple transformation results according to a preset feature template to obtain multiple audio features; Calculating the feature similarity between each audio feature and a preset feature, and determining the audio feature with the greatest feature similarity as the user audio feature; Counting the voice duration of the first user in each round of conversation information according to the user audio feature as the audio duration; Determine the round in which the audio duration is greater than the preset duration as the target round; The key audio is extracted from the audio to be inspected according to the target round.

3. The speech quality inspection method based on artificial intelligence according to claim 1, characterized in that: The extracting of speech feature parameters in the key audio based on the speech processing network comprises: Obtaining a transfer function and a windowing function from the speech processing network; Performing pre-emphasis processing on the key audio based on the transfer function to obtain a first audio; Performing frame processing on the first audio to obtain multiple speech frames; Obtaining a frame element of each speech frame, and calculating the product of the frame element and the windowing function to obtain a spectrogram; The spectrogram is input into a convolutional feature extraction layer in the speech processing network to obtain the speech feature parameters.

4. The method for speech quality inspection based on artificial intelligence according to claim 1, characterized in that: The emotion processing network includes a tone recognition layer and a semantic analysis layer, and the speech feature parameters and the key audio are recognized based on the emotion processing network to obtain the target emotion, including: Recognize the speech feature parameters based on the tone recognition layer to obtain tone information; Convert the key audio into text information, and parse the text information based on the semantic parsing layer to obtain semantic information; Acquire a first emotion corresponding to the tone information, and acquire a second emotion corresponding to the semantic information; If the first emotion is the same as the second emotion, determining the first emotion or the second emotion as the target emotion; or If the first emotion is different from the second emotion, obtain the tone score corresponding to the tone information and the semantic score corresponding to the semantic information, weight and process the tone score and the semantic score to obtain the emotion score, and determine the type corresponding to the emotion score as the target emotion.

5. The artificial intelligence-based speech quality inspection method according to claim 4, characterized in that: The performing speech speed detection on the target audio to obtain the speech speed detection result of the second user includes: Extracting information corresponding to the target audio from the text information as target text; Counting the text volume of the target text, and counting the duration of the target audio as the target duration; The ratio of the text amount to the target duration is calculated to obtain the speech rate detection result.

6. A speech quality inspection device based on artificial intelligence, characterized in that: The speech quality inspection device based on artificial intelligence includes: an acquisition unit, configured to acquire the audio to be inspected according to the quality inspection request when a quality inspection request is received, wherein the audio to be inspected includes information about multiple rounds of conversations between the first user and the second user; An extraction unit, configured to extract key audio from the audio to be inspected according to the audio duration of the first user in the multi-round conversation information and a preset duration; The acquisition unit is further used to acquire a pre-trained speech emotion recognition model, wherein the speech emotion recognition model includes a speech processing network and an emotion processing network; The extraction unit is further used to extract speech feature parameters in the key audio based on the speech processing network; A recognition unit, configured to recognize the speech feature parameters and the key audio based on the emotion processing network to obtain a target emotion; A removing unit, used for removing the conversation information of the first user from the key audio to obtain the target audio; A detection unit, configured to perform silence detection on the target audio to obtain a silence detection result of the second user, comprising: obtaining a plurality of target frames from the target audio; extracting a target speech feature from each target frame; inputting the target speech feature into a pre-trained silence classifier to obtain a category result for each target frame; detecting a speech occurrence frame and a speech disappearance frame in the target audio according to the category result and the timing information of each target frame in the target audio; calculating the number of distance frames between the speech disappearance frame and the speech occurrence frame; and calculating the ratio of the number of distance frames to the number of the plurality of target frames to obtain the silence detection result; The detection unit is further configured to perform speech speed detection on the target audio to obtain a speech speed detection result of the second user; A generating unit, configured to generate a quality inspection result according to the target emotion, the silence detection result, and the speech speed detection result, comprising: obtaining a score interval corresponding to the target emotion, the score interval comprising a first score and a second score, the first score being less than the second score; and calculating the first score, the silence detection result, and the speech speed detection result according to the following formula to obtain first quality inspection information: Among them, y refers to the first quality inspection information, h refers to the first score, f refers to the silence detection result, and x refers to the speech rate detection result; the second score, the silence detection result and the speech rate detection result are analyzed to obtain the second quality inspection information; the quality inspection result is generated according to the first quality inspection information and the second quality inspection information.

7. An electronic device, characterized in that: The electronic device comprises: a memory storing computer-readable instructions; and A processor executes computer-readable instructions stored in the memory to implement the artificial intelligence-based speech quality inspection method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions are executed by a processor in an electronic device to implement the artificial intelligence-based speech quality inspection method as described in any one of claims 1 to 5.