Speech synthesis parameter adaptive adjustment method, device, computer equipment and storage medium

Abstract

Embodiments of the present invention disclose a method, apparatus, computer device, and storage medium for adaptively adjusting speech synthesis parameters. The method comprises: organizing marketing speech knowledge; obtaining customer information and conversation information from outbound calls, and generating initial outbound speech synthesis parameters based on the customer information and conversation information; obtaining actual conversation content; analyzing the adjustment intent within the conversation scenario based on the actual conversation content; generating real-time speech synthesis parameters based on the adjustment intent and the initial outbound speech synthesis parameters; and generating real-time outbound speech based on the real-time speech synthesis parameters. Implementing the method of the embodiments of the present invention allows for adjusting speech synthesis parameters based on the actual conversation scenario, thereby improving the ultimate telemarketing conversion rate.

Description

Technical Field

[0001] The present invention relates to a speech synthesis method, and more specifically to a speech synthesis parameter adaptive adjustment method, device, computer equipment and storage medium. Background Art

[0002] In outbound marketing scenarios, for example, in the banking sector, marketing products mainly include credit cards, large deposits and loans, and wealth management products. Due to the large number of marketing products and a large customer base, traditional intelligent outbound call systems generally set up voice synthesis parameter information in advance. As a result, when conducting product marketing, the system cannot automatically adjust the speed and urgency of the outbound call according to the caller and the conversation scenario at the time, which ultimately affects the marketing effect.

[0003] Therefore, it is necessary to design a new method to adjust the speech synthesis parameters according to the actual conversation scenario and improve the final telephone marketing conversion rate. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the prior art and provide a method, device, computer equipment and storage medium for adaptively adjusting speech synthesis parameters.

[0005] To achieve the above object, the present invention adopts the following technical solution: a method for adaptively adjusting speech synthesis parameters, comprising:

[0006] Organize marketing skills knowledge;

[0007] Obtaining customer information and conversation information of an outbound call, and generating initial parameters for outbound speech synthesis based on the customer information and conversation information;

[0008] Get the actual conversation content;

[0009] analyzing the adjustment intention in the dialogue scenario based on the actual dialogue content;

[0010] generating real-time speech synthesis parameters according to the adjustment intention and the initial parameters for outbound speech synthesis;

[0011] Generate real-time outbound voice according to the real-time voice synthesis parameters.

[0012] Its further technical solution is: the said organizing marketing speech knowledge includes:

[0013] Organizing graphic materials of graphic and text introduction documents of products to be marketed, organizing basic information of products to be marketed according to the graphic and text materials, and converting the basic information of products to be marketed into the form of questions and answers for storage;

[0014] Organize the basic information of marketing promoters and convert it into questions and answers for storage;

[0015] Organize abnormal conversation words.

[0016] A further technical solution is: the step of obtaining customer information and conversation information of an outbound call, and generating initial parameters for outbound speech synthesis based on the customer information and conversation information, includes:

[0017] Obtain customer information and conversation information for outbound calls;

[0018] Predicting user connection quality of different speech synthesis parameters using a GBDT model based on the customer information and the conversation information to obtain predicted values ​​of the different speech synthesis parameters;

[0019] The speech synthesis parameter with the largest predicted value among different speech synthesis parameters is selected to obtain the initial parameter for outbound speech synthesis.

[0020] A further technical solution is: analyzing the adjustment intention in the dialogue scene according to the actual dialogue content includes:

[0021] analyzing explicit voice adjustment intentions within the actual conversation content;

[0022] analyzing implicit voice adjustment intentions within the actual conversation content;

[0023] The explicit voice adjustment intention and the implicit voice adjustment intention are integrated to form an adjustment intention in a dialogue scenario.

[0024] A further technical solution is: analyzing the explicit voice adjustment intention in the actual conversation content includes:

[0025] Using TextCNN to perform short text classification on the actual conversation content to obtain explicit voice adjustment intentions;

[0026] The short text classification of the actual conversation content using TextCNN to obtain the explicit voice adjustment intention includes:

[0027] Segmenting the speech text corresponding to the actual conversation content to obtain words;

[0028] Vectorizing the words to obtain word vectors;

[0029] Perform convolution feature mapping on the word vector using N convolution kernels to obtain semantic convolution feature information;

[0030] Perform maximum pooling on the semantic convolution feature information to obtain the pooling value;

[0031] The pooled values ​​are passed through a fully connected softmax layer for length compression to obtain the probability of the category;

[0032] Determine explicit speech adjustment intention based on category probabilities.

[0033] A further technical solution is: the analysis of the implicit voice adjustment intention in the actual conversation content includes:

[0034] Classifying the actual conversation content using a CNN model to obtain implicit voice adjustment intentions;

[0035] The classifying the actual conversation content using a CNN model to obtain implicit voice adjustment intentions includes:

[0036] Segmenting the speech text of the actual conversation content whose context length is within a set range to obtain a plurality of text segmentations;

[0037] Sequentially vectorizing the text segmentation words to obtain a plurality of word vectors;

[0038] Performing convolution feature mapping on a plurality of the word vectors in sequence to obtain a plurality of semantic convolution feature information;

[0039] Calculating vector differences and vector difference ratios between the plurality of semantic convolution feature information in a pairwise combination manner, and performing convolution feature extraction on the vector difference ratios again to obtain similarity convolution features and connectivity convolution features between the vectors;

[0040] The similarity convolutional features and the connectivity convolutional features are subjected to maximum pooling and length compression through a fully connected softmax layer to obtain the probability of the category;

[0041] Determining implicit speech adjustment intentions based on class probabilities.

[0042] A further technical solution is: generating real-time outbound voice according to the real-time voice synthesis parameters includes:

[0043] The real-time speech synthesis parameters are sent to a speech synthesis model, so that the speech synthesis model generates real-time outbound speech according to the real-time speech synthesis parameters.

[0044] The present invention also provides a speech synthesis parameter adaptive adjustment device, comprising:

[0045] Knowledge organization unit, used to organize marketing knowledge;

[0046] An initial parameter generation unit, configured to obtain customer information and conversation information of an outbound call, and generate initial parameters for outbound speech synthesis based on the customer information and conversation information;

[0047] A conversation content acquisition unit, used to acquire actual conversation content;

[0048] an intention analysis unit, configured to analyze the adjustment intention in the dialogue scenario based on the actual dialogue content;

[0049] A real-time parameter generating unit, configured to generate real-time speech synthesis parameters according to the adjustment intention and the outbound speech synthesis initial parameters;

[0050] The speech synthesis unit is used to generate real-time outbound speech according to the real-time speech synthesis parameters.

[0051] The present invention further provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0052] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0053] The beneficial effects of the present invention compared with the existing technology are: the present invention organizes marketing speech knowledge and generates initial parameters for outbound speech synthesis based on customer information and conversation information. During the actual conversation process, the conversation content is analyzed in real time and the adjustment intention is determined. The initial parameters for outbound speech synthesis are adjusted to generate real-time speech synthesis parameters, and outbound speech is generated from the real-time speech synthesis parameters, thereby adjusting the speech synthesis parameters according to the actual conversation scenario and improving the final telephone marketing conversion rate.

[0054] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 A schematic diagram of an application scenario of the method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0057] Figure 2 A schematic diagram of a flow chart of a method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0058] Figure 3 A schematic diagram of a sub-process of a method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0059] Figure 4 A schematic diagram of a sub-process of a method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0060] Figure 5 A schematic diagram of a sub-process of a method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0061] Figure 6 A schematic diagram of a sub-process of a method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0062] Figure 7 A schematic diagram of a sub-process of a method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0063] Figure 8 A schematic diagram of call quality scores for the method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0064] Figure 9 A schematic block diagram of a device for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0065] Figure 10 A schematic block diagram of a knowledge collating unit of the apparatus for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0066] Figure 11 A schematic block diagram of an initial parameter generation unit of the speech synthesis parameter adaptive adjustment device provided by an embodiment of the present invention;

[0067] Figure 12 A schematic block diagram of an intention analysis unit of the apparatus for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention;

[0068] Figure 13 A schematic block diagram of an explicit intention analysis subunit of the speech synthesis parameter adaptive adjustment device provided by an embodiment of the present invention;

[0069] Figure 14 A schematic block diagram of an implicit intention analysis subunit of the speech synthesis parameter adaptive adjustment device provided by an embodiment of the present invention;

[0070] Figure 15 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0072] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0073] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0074] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0075] See also Figure 1 and Figure 2 , Figure 1 A schematic diagram of an application scenario of the method for adaptively adjusting speech synthesis parameters provided by an embodiment of the present invention. Figure 2 This is a schematic flow chart of a method for adaptively adjusting speech synthesis parameters provided in an embodiment of the present invention. This method is applied to a server. The server interacts with outbound call devices, such as intelligent robots, and speech synthesis devices. The server determines initial parameters by organizing marketing techniques and obtaining customer and conversation information from outbound calls. The server then obtains actual conversation content from the intelligent robot or other outbound call device, determines the intended adjustment to the conversation process based on that content, and then adjusts speech synthesis parameters, such as volume and speech rate, in real time based on the intended adjustment to determine the final speech synthesis parameters. The speech synthesis device then performs speech synthesis.

[0076] Figure 2 FIG. 1 is a flow chart of a method for adaptively adjusting speech synthesis parameters according to an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S160.

[0077] S110. Organize marketing language knowledge.

[0078] In this embodiment, the marketing language knowledge includes basic product information, product life, profitability introduction, risk introduction, target customer groups, purchase process, maturity or early redemption and other document introductions, the promotion and marketing person's own situation introduction, etc., as well as abnormal conversation situations.

[0079] Specifically, the products to be marketed refer to banking products, primarily including credit cards, large-value deposits and loans, and wealth management products. Promotional materials for each product need to be compiled, primarily including basic product information, product lifespan, profitability, risk profile, target customer base, purchase process, and documentation regarding maturity or early redemption. Secondly, the script should also include an introduction to the marketer's role. Finally, due to the use of smartphones for outbound calls, preparations need to be made for unusual conversation situations, such as muting, unclear speech, background noise, requests for repetition, and polite greetings.

[0080] In one embodiment, see Figure 3 , the above-mentioned step S110 may include steps S111 to S113.

[0081] S111 , organizing graphic materials of a graphic and text introduction document of the product to be marketed, organizing basic information of the product to be marketed according to the graphic and text materials, and converting the basic information of the product to be marketed into a form of questions and answers for storage.

[0082] Specifically, organize the graphic and text materials of the product to be marketed, and organize the basic information of the product to be marketed according to the graphic and text materials. The basic information of the product to be marketed includes the product name, issuing entity, issuance scope, product life, profitability introduction, risk introduction, target customer group, purchase process, maturity or early redemption, and other basic information descriptions. Generally speaking, the basic feature information of the product is usually stored in PDF documents or the column of the homepage product page, and some are presented in the form of tables or pictures. Therefore, it is necessary to sort out this information according to the basic items of product information mentioned above, convert it into questions and answers for storage, and finally turn it into knowledge material for the marketing robot, which is the backbone information of the entire dialogue process.

[0083] S112. Organize the basic information of the marketing promoters and convert the basic information of the marketing promoters into questions and answers for storage.

[0084] In this embodiment, the basic information of the marketing promoter includes basic information such as the current bank, bank address, bank customer service phone number, and bank website.

[0085] Specifically, the basic information of the marketing promoters represented by the intelligent outbound call robot is sorted out and converted into the form of questions and answers.

[0086] S113. Organize abnormal conversation words.

[0087] In this embodiment, abnormal conversational speech mainly includes speech corresponding to situations such as the other party is busy, the other party is complaining, the other party is muted, the other party's voice is unclear, the other party repeats the question, the other party actively asks to say it again, the other party's background is noisy, it is not the real person, and polite greetings.

[0088] Specifically, abnormal dialogue language may be repeatedly interspersed throughout the entire marketing scenario, and it is also a place where special attention needs to be paid when soothing sudden emotions during conversations. How well these situations are handled is crucial to the final marketing conversion and avoiding customer complaints.

[0089] By organizing the above three aspects, we can acquire the marketing scripting knowledge for the entire conversational robot. Generally, for the basic product text, we need to prepare three versions: a simplified version, a moderate version, and a detailed Q&A version. During speech synthesis, we can also identify the caller's emotion and answering status to control the robot's selection of more appropriate scripting content and control its speech speed, emphasis, etc.

[0090] S120: Obtain customer information and conversation information of the outbound call, and generate initial parameters for outbound speech synthesis based on the customer information and conversation information.

[0091] In this embodiment, the initial parameters for outbound speech synthesis refer to parameter information such as timbre, speed, volume, and pitch that are initially selected based on information such as the user's gender, age, and occupation.

[0092] In one embodiment, see Figure 4 , the above-mentioned step S120 may include steps S121 to S123.

[0093] S121. Obtain customer information and conversation information of the outbound call.

[0094] In this embodiment, customer information refers to customer personalized information and user information such as gender, age, and occupation; conversation information refers to the current outbound call time and product information of the current outbound marketing.

[0095] S122 . Predict user connection quality of different speech synthesis parameters using a GBDT model according to the customer information and the conversation information, so as to obtain predicted values ​​of the different speech synthesis parameters.

[0096] In this embodiment, the predicted values ​​of different speech synthesis parameters refer to scores obtained by using different speech parameters such as different timbre, speed, volume, pitch and other parameter information as initial parameters to predict user connection quality.

[0097] S123. Filter the speech synthesis parameter with the largest predicted value among different speech synthesis parameters to obtain the initial parameter for outbound speech synthesis.

[0098] Specifically, before an outbound call robot makes an outbound call to a customer, it needs to preset the conversation tone, speed and other parameter information based on the customer's personalized information. It mainly selects different tone, speed, volume, tone and other parameter information based on the user's gender, age, occupation and other information, and selects the appropriate marketing opening remarks based on the current user information.

[0099] In this embodiment, timbre is categorized into several common voice types, such as standard male and standard female voices. Speed ​​is divided into 10 levels based on the number of words read per minute, from low to high. Volume and pitch are also categorized into 10 levels based on the decibel level and pitch of the human voice. This makes it a matter of selecting the optimal initial parameters for outbound speech synthesis based on the user's current information.

[0100] The user's age, gender, occupation, work schedule, and other factors, as well as the current outbound call time and product information for the current outbound marketing campaign, largely determine the appropriate timbre, pitch, speaking speed, and volume for the current outbound call. Therefore, this problem can be considered a classification problem, where the optimal speech synthesis parameters are determined based on this characteristic information. What are the optimal speech synthesis parameters? We use historically answered calls for prediction. We ignore calls that were not connected or were directly hung up, as they did not constitute actual conversations. We then categorize these answered call records into different quality levels, with the highest score for call intent and the highest score for more conversation turns. Call quality scores are divided into six levels, from low to high, as shown in Table 1 below.

[0101] Table 1. Call quality scores

[0102]

[0103] Therefore, selecting the best speech synthesis parameters becomes a matter of predicting the final call quality score for the current user, current scenario, and current time under different speech synthesis parameters, and selecting the highest predicted score. Figure 8 As shown in the figure, under the conditions of user determination, outbound call scenario characteristics, and current outbound call time, appropriate speech synthesis parameter characteristics are selected to maximize the final call quality score. For this problem, the user's age, gender, occupation, work schedule, etc. are discretely labeled, and then the level characteristics of different timbre, pitch, volume, and speed are adjusted to predict the final call quality score, and the maximized predicted score is selected.

[0104] The GBDT (Gradient Boosting Decision Tree) model is used to predict user connection quality, primarily due to its excellent performance in classification and linear fitting. GBDT is a type of boosting model that constructs multiple CART regression trees. Each round of GBDT training is based on the residuals from the previous round of training, as well as the negative gradient values ​​of the current model. During GBDT classification training, a classification regression tree is trained for each class of sample. As shown in the score evaluation above, there are six categories for user connection quality ratings. Therefore, six trees are trained simultaneously in each round of training, with each tree trained specifically for a specific type of data. This determines the optimal speech synthesis parameter information for the initialization phase of an outbound call.

[0105] S130: Obtain actual conversation content.

[0106] In this embodiment, the actual conversation content refers to the conversation text collected during the actual outbound call process.

[0107] S140: Analyze the adjustment intention in the dialogue scenario according to the actual dialogue content.

[0108] In this embodiment, the adjustment intention refers to the adjustment intention for speech synthesis parameters such as volume and speech speed.

[0109] During an outbound call, it may be necessary to monitor and adjust voice synthesis parameters such as volume and speech speed in real time during the call due to the location and environment of the called person, the matter they are handling, or the fact that they are not the original customer, or the customer did not hear clearly, or actively asked to speak slower, faster, louder or softer.

[0110] In one embodiment, see Figure 5 , the above-mentioned step S140 may include steps S141 to S143.

[0111] S141: Analyze the explicit voice adjustment intention in the actual conversation content.

[0112] In this embodiment, the explicit voice adjustment intention refers to the intention of adjusting voice synthesis parameters such as volume and speech speed that is clearly indicated by the user during the conversation.

[0113] Specifically, TextCNN is used to perform short text classification on the actual conversation content to obtain explicit voice adjustment intentions.

[0114] In one embodiment, see Figure 6 , the above-mentioned step S141 may include steps S1411 to S1416.

[0115] S1411. Segment the speech text corresponding to the actual conversation content to obtain words;

[0116] S1412: Vectorize the words to obtain word vectors.

[0117] S1413, performing convolution feature mapping on the word vector using N convolution kernels to obtain semantic convolution feature information;

[0118] S1414, performing maximum pooling on the semantic convolution feature information to obtain a pooling value;

[0119] S1415, compressing the length of the pooled values ​​through a fully connected softmax layer to obtain the probability of the category;

[0120] S1416. Determine the explicit voice adjustment intention based on the category probability.

[0121] Specifically, users may require adjustment of their voice during a call due to various reasons. Figure 1 Generally, there are two situations: explicit voice adjustment intention and implicit voice adjustment intention. Explicit voice adjustment intention is when the user explicitly asks to adjust the voice louder or softer, or to speak faster or slower. For the recognition of explicit voice adjustment intention, short text classification is used for recognition. Since the timbre of speech synthesis has been determined in the initialization stage, this speech synthesis parameter cannot be changed. The displayed voice adjustment during the call process can be divided into five categories: increase volume, decrease volume, increase speaking speed, decrease speaking speed, and others. Therefore, this technology can identify the current user's speech intention, specifically using TextCNN short text classification.

[0122] First, the user's speech text is segmented and converted into basic words; the words are vectorized and represented, and each word is represented by a 256-dimensional Word2Vec word vector. Therefore, the user's speech can be represented as a two-dimensional vector of [Sen_Length, Embedding_Dim], where Sen_Length is the sentence length and Embedding_Dim is the vectorized representation of each segmented word, which is 256 here. N convolution kernels are used to obtain N convolution feature maps, and the semantic convolution feature information in the user's speech text is obtained; max-pooling is performed on the convolution feature map to obtain N 1x1 values, so that sentences of different lengths can be converted into fixed-length representations after the pooling layer, and then spliced ​​into an N-dimensional vector as the sentence representation of the text; finally, a fully connected softmax layer is connected to compress the N-dimensional vector to a 1-dimensional vector of length 5 (this is because the user's displayed adjustment has a total of 5 categories), and the probability of each category is output, thereby obtaining the intention of adjusting the explicit voice.

[0123] S142: Analyze the implicit voice adjustment intention in the actual conversation content.

[0124] In this embodiment, implicit voice adjustment intent refers to detecting implicit voice adjustment intent. For example, when a user attempts to attract the other party's attention by saying "Hello, hello," repeating a previous question, or explaining something like "It's a bit noisy," the other party is implicitly suggesting that the speed or volume should be adjusted. For implicit feedback, repetitiveness is typically calculated by calculating the similarity of user contextual speech, or by analyzing whether the user understood the previous question based on the context of the question and response.

[0125] In one embodiment, see Figure 7 , the above-mentioned step S142 may include steps S1421 to S1426.

[0126] S1421. Segment the speech texts whose context lengths of the actual conversation content are within a set range to obtain a plurality of text segmentations;

[0127] S1422. Sequentially vectorize the plurality of text segmentations to obtain a plurality of word vectors.

[0128] S1423, performing convolution feature mapping on the plurality of word vectors in sequence to obtain a plurality of semantic convolution feature information;

[0129] S1424. Calculate the vector difference and vector difference ratio between the plurality of semantic convolution feature information in a pairwise combination manner, and perform convolution feature extraction on the vector difference ratio again to obtain similarity convolution features and connectivity convolution features between the vectors.

[0130] S1425, performing maximum pooling on the similarity convolutional features and the connectivity convolutional features, and performing length compression through a fully connected softmax layer to obtain category probabilities;

[0131] S1426. Determine the implicit speech adjustment intention based on the category probability.

[0132] Generally speaking, implicit voice adjustments are performed when the user is not listening clearly. Similar to the analysis of explicit voice adjustment intent described above, a CNN model is also used for classification. This model primarily models the overlap and cohesion of the user's previous question, the phone question, and the user's current response (a context length of three). This determines whether there are repeated questions or irrelevant answers. This determines the parameters for adjusting speech synthesis. Specifically, the conversation text within a call context length of three is first segmented and converted into basic words. These three text segments are then vectorized, with each word represented by a 256-dimensional Word2Vec word vector. These three vectors are then convolved with N kernels to generate N convolutional feature maps, which represent the semantic convolutional feature information of the user's conversation text. The vector difference and vector difference ratio are then calculated pairwise between the three text vector convolutional features. These three difference features are then further convolved to extract features, resulting in convolutional features that measure similarity and cohesion between the vectors. Max-pooling is performed on the convolutional feature map, and finally a fully connected softmax layer is added to compress the N-dimensional vector into a 1-dimensional vector of length 5. This is because the user's displayed adjustments have five categories, and the probability of each category is output. This way, the implicit voice adjustment intention is obtained.

[0133] S143: Integrate the explicit voice adjustment intention and the implicit voice adjustment intention to form an adjustment intention in the dialogue scenario.

[0134] S150: Generate real-time speech synthesis parameters according to the adjustment intention and the initial parameters for outbound speech synthesis.

[0135] In this embodiment, the real-time speech synthesis parameters refer to parameter results formed by adjusting the initial parameters of the outbound speech synthesis according to the adjustment intention.

[0136] The initial outbound speech synthesis parameters display the initial timbre, tempo, pitch, and volume parameters for a call. In-call voice adjustment recognition then determines the user's intention to adjust these parameters, such as volume and tempo. If the user intends to adjust, the system adjusts the volume, tempo, and pitch levels in the finest possible steps based on the intended adjustment. Fine-tuning is performed on the current latest speech synthesis parameters by ±1 level, resulting in the latest speech synthesis parameters.

[0137] S160: Generate real-time outbound voice according to the real-time voice synthesis parameters.

[0138] Specifically, the real-time speech synthesis parameters are sent to a speech synthesis model, so that the speech synthesis model generates real-time outbound speech according to the real-time speech synthesis parameters.

[0139] Real-time speech synthesis parameters are used as hyperparameters and integrated into the speech synthesis model. Since outbound calls are typically made using third-party open voice platforms like Baidu and Alibaba, this type of speech synthesis can be directly adjusted through the open API call layer to obtain the adjusted speech synthesis variables.

[0140] For the self-developed speech synthesis model, we need to pass the speech synthesis parameters to the speech synthesis model through the duration model of the synthesis model. The duration model mainly controls the pronunciation duration of the speech notes, the pause duration between notes and other information. It can generally be used to generate state duration parameters in speech synthesis based on the hidden Markov model. For the input context-related hidden Markov model sequence, the duration of each state of each model is generated, that is, the residence time of each state; thus, real-time outbound voice is formed.

[0141] By organizing telephone voice text materials, generating initialization parameters for outbound voice calls, and adjusting voice synthesis parameters during outbound calls, we can effectively determine the voice synthesis parameters during outbound calls and adaptively adjust them in real time, thereby improving the smoothness and interactivity of the interaction, and ultimately enhancing the marketing experience and conversion.

[0142] The above-mentioned method for adaptively adjusting speech synthesis parameters organizes marketing speech knowledge and generates initial parameters for outbound speech synthesis based on customer information and conversation information. During the actual conversation process, the conversation content is analyzed in real time and the adjustment intention is determined. The initial parameters for outbound speech synthesis are adjusted to generate real-time speech synthesis parameters, and outbound speech is generated from the real-time speech synthesis parameters. This adjusts the speech synthesis parameters according to the actual conversation scenario and improves the final telephone marketing conversion rate.

[0143] Figure 9 FIG is a schematic block diagram of a speech synthesis parameter adaptive adjustment device 300 provided by an embodiment of the present invention. Figure 9 As shown, corresponding to the above speech synthesis parameter adaptive adjustment method, the present invention also provides a speech synthesis parameter adaptive adjustment device 300. The speech synthesis parameter adaptive adjustment device 300 includes a unit for executing the above speech synthesis parameter adaptive adjustment method, and the device can be configured in a server. Figure 9 The speech synthesis parameter adaptive adjustment device 300 includes a knowledge organization unit 301, an initial parameter generation unit 302, a dialogue content acquisition unit 303, an intention analysis unit 304, a real-time parameter generation unit 305 and a speech synthesis unit 306.

[0144] The knowledge organization unit 301 is used to organize marketing speech knowledge; the initial parameter generation unit 302 is used to obtain customer information and dialogue information of the outbound conversation, and generate initial parameters for outbound speech synthesis based on the customer information and dialogue information; the dialogue content acquisition unit 303 is used to obtain the actual dialogue content; the intention analysis unit 304 is used to analyze the adjustment intention in the dialogue scenario based on the actual dialogue content; the real-time parameter generation unit 305 is used to generate real-time speech synthesis parameters based on the adjustment intention and the outbound speech synthesis initial parameters; the speech synthesis unit 306 is used to generate real-time outbound speech based on the real-time speech synthesis parameters.

[0145] In one embodiment, if Figure 10 As shown, the knowledge arranging unit 301 includes a first arranging subunit 3011 , a second arranging subunit 3012 and a third arranging subunit 3013 .

[0146] The first arranging subunit 3011 is used to arrange the graphic materials of the graphic and text introduction document of the product to be marketed, and to arrange the basic information of the product to be marketed according to the graphic and text materials, and to convert the basic information of the product to be marketed into the form of questions and answers for storage; the second arranging subunit 3012 is used to arrange the basic information of the marketing promoter, and to convert the basic information of the marketing promoter into the form of questions and answers for storage; the third arranging subunit 3013 is used to arrange abnormal dialogue words.

[0147] In one embodiment, if Figure 11 As shown, the initial parameter generation unit 302 includes an information acquisition subunit 3021 , a prediction subunit 3022 and a screening subunit 3023 .

[0148] The information acquisition subunit 3021 is used to obtain customer information and conversation information of the outbound call conversation; the prediction subunit 3022 is used to use the GBDT model to predict the user connection quality of different speech synthesis parameters based on the customer information and the conversation information to obtain the predicted values ​​of the different speech synthesis parameters; the screening subunit 3023 is used to screen the speech synthesis parameter with the largest predicted value among the different speech synthesis parameters to obtain the initial parameters for the outbound speech synthesis.

[0149] In one embodiment, if Figure 12 As shown, the intention analysis unit 304 includes an explicit intention analysis subunit 3041 , an implicit intention analysis subunit 3042 and an integration subunit 3043 .

[0150] The explicit intention analysis subunit 3041 is used to analyze the explicit voice adjustment intention in the actual conversation content; the implicit intention analysis subunit 3042 is used to analyze the implicit voice adjustment intention in the actual conversation content; the integration subunit 3043 is used to integrate the explicit voice adjustment intention and the implicit voice adjustment intention to form the adjustment intention in the conversation scene.

[0151] In one embodiment, the explicit intention analysis subunit 3041 is used to perform short text classification on the actual conversation content using TextCNN to obtain the explicit voice adjustment intention.

[0152] In one embodiment, if Figure 13 As shown, the explicit intent analysis subunit 3041 includes a first word segmentation module 30411, a first vectorization module 30412, a first mapping module 30413, a first pooling module 30414, a first compression module 30415 and a first determination module 30416.

[0153] The first word segmentation module 30411 is used to segment the speech text corresponding to the actual conversation content to obtain words; the first vectorization module 30412 is used to vectorize the words to obtain word vectors. The first mapping module 30413 is used to perform convolution feature mapping on the word vector using N convolution kernels to obtain semantic convolution feature information; the first pooling module 30414 is used to perform maximum pooling on the semantic convolution feature information to obtain pooled values; the first compression module 30415 is used to perform length compression on the pooled values ​​through a fully connected softmax layer to obtain category probabilities; the first determination module 30416 is used to determine the explicit voice adjustment intention based on the category probabilities.

[0154] In one embodiment, the implicit intention analysis subunit 3042 is used to classify the actual conversation content using a CNN model to obtain implicit voice adjustment intention.

[0155] In one embodiment, if Figure 14 As shown, the implicit intent analysis subunit 3042 includes a second word segmentation module 30421, a second vectorization module 30422, a second mapping module 30423, an extraction module 30424, a processing module 30425 and a second determination module 30426.

[0156] The second word segmentation module 30421 is used to segment the speech text whose context length of the actual dialogue content is within a set range to obtain a number of text segmentations; the second vectorization module 30422 is used to vectorize the plurality of text segmentations in turn to obtain a number of word vectors; the second mapping module 30423 is used to perform convolution feature mapping on the plurality of word vectors in turn to obtain a number of semantic convolution feature information; the extraction module 30424 is used to calculate the vector difference and the vector difference ratio between the plurality of semantic convolution feature information in a pairwise manner, and perform convolution feature extraction on the vector difference ratio again to obtain similarity convolution features and connectivity convolution features between the vectors; the processing module 30425 is used to perform maximum pooling on the similarity convolution features and connectivity convolution features, and perform length compression through a fully connected softmax layer to obtain the probability of the category; the second determination module 30426 is used to determine the implicit speech adjustment intention based on the category probability.

[0157] In one embodiment, the speech synthesis unit 306 is configured to send the real-time speech synthesis parameters to a speech synthesis model, so that the speech synthesis model generates real-time outbound speech according to the real-time speech synthesis parameters.

[0158] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned speech synthesis parameter adaptive adjustment device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment. For the convenience and brevity of description, it will not be repeated here.

[0159] The above-mentioned speech synthesis parameter adaptive adjustment device 300 can be implemented in the form of a computer program. The computer program can be used in Figure 15 Runs on the computer equipment shown.

[0160] See also Figure 15 , Figure 15 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0161] See Figure 15 The computer device 500 includes a processor 502 , a memory, and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0162] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, which, when executed, can enable the processor 502 to execute a method for adaptively adjusting speech synthesis parameters.

[0163] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0164] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a method for adaptively adjusting speech synthesis parameters.

[0165] The network interface 505 is used to communicate with other devices through the network. Figure 15 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0166] The processor 502 is configured to execute a computer program 5032 stored in the memory to implement the following steps:

[0167] Organize marketing speech knowledge; obtain customer information and conversation information of outbound conversations, and generate initial parameters for outbound speech synthesis based on the customer information and conversation information; obtain actual conversation content; analyze adjustment intentions in conversation scenarios based on the actual conversation content; generate real-time speech synthesis parameters based on the adjustment intentions and the initial parameters for outbound speech synthesis; generate real-time outbound speech based on the real-time speech synthesis parameters.

[0168] In one embodiment, when implementing the step of organizing marketing speech knowledge, the processor 502 specifically implements the following steps:

[0169] Organize the graphic and text materials of the graphic and text introduction documents of the products to be marketed, and organize the basic information of the products to be marketed according to the graphic and text materials, and convert the basic information of the products to be marketed into the form of questions and answers for storage; organize the basic information of the marketing promoters, and convert the basic information of the marketing promoters into the form of questions and answers for storage; organize abnormal conversation words.

[0170] In one embodiment, when the processor 502 implements the step of obtaining the customer information and conversation information of the outbound call conversation and generating the initial parameters for outbound speech synthesis based on the customer information and conversation information, the processor 502 specifically implements the following steps:

[0171] Obtain customer information and conversation information of an outbound call conversation; use the GBDT model to predict user connection quality of different speech synthesis parameters based on the customer information and the conversation information to obtain predicted values ​​of the different speech synthesis parameters; and select the speech synthesis parameter with the largest predicted value of the different speech synthesis parameters to obtain the initial parameters for the outbound speech synthesis.

[0172] In one embodiment, when the processor 502 implements the step of analyzing the adjustment intention in the conversation scenario according to the actual conversation content, it specifically implements the following steps:

[0173] Analyze the explicit voice adjustment intention in the actual conversation content; analyze the implicit voice adjustment intention in the actual conversation content; integrate the explicit voice adjustment intention and the implicit voice adjustment intention to form the adjustment intention in the conversation scene.

[0174] In one embodiment, when implementing the step of analyzing the explicit voice adjustment intention in the actual conversation content, the processor 502 specifically implements the following steps:

[0175] TextCNN is used to perform short text classification on the actual conversation content to obtain explicit voice adjustment intentions.

[0176] In one embodiment, when implementing the step of using TextCNN to perform short text classification on the actual conversation content to obtain the explicit voice adjustment intention, the processor 502 specifically implements the following steps:

[0177] The speech text corresponding to the actual conversation content is segmented to obtain words; the words are vectorized to obtain word vectors; the word vectors are convolutionally mapped using N convolution kernels to obtain semantic convolution feature information; the semantic convolution feature information is maximum pooled to obtain a pooled value; the pooled value is length-compressed through a fully connected softmax layer to obtain a category probability; and the explicit voice adjustment intention is determined based on the category probability.

[0178] In one embodiment, when implementing the step of analyzing the implicit voice adjustment intention in the actual conversation content, the processor 502 specifically implements the following steps:

[0179] The actual conversation content is classified using a CNN model to obtain implicit voice adjustment intentions.

[0180] In one embodiment, when implementing the step of classifying the actual conversation content using a CNN model to obtain the implicit voice adjustment intention, the processor 502 specifically implements the following steps:

[0181] Segment the speech text within a set context length of the actual conversation content to obtain a number of text segmentations; vectorize the text segmentations in turn to obtain a number of word vectors; perform convolution feature mapping on the word vectors in turn to obtain a number of semantic convolution feature information; calculate the vector difference and vector difference ratio between the semantic convolution feature information in a pairwise manner, and perform convolution feature extraction on the vector difference ratio again to obtain similarity convolution features and connectivity convolution features between the vectors; perform maximum pooling on the similarity convolution features and connectivity convolution features, and perform length compression through a fully connected softmax layer to obtain category probability; determine the implicit speech adjustment intention based on the category probability.

[0182] In one embodiment, when implementing the step of generating the real-time outbound voice according to the real-time voice synthesis parameters, the processor 502 specifically implements the following steps:

[0183] The real-time speech synthesis parameters are sent to a speech synthesis model, so that the speech synthesis model generates real-time outbound speech according to the real-time speech synthesis parameters.

[0184] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be 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. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0185] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0186] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor performs the following steps:

[0187] Organize marketing speech knowledge; obtain customer information and conversation information of outbound conversations, and generate initial parameters for outbound speech synthesis based on the customer information and conversation information; obtain actual conversation content; analyze adjustment intentions in conversation scenarios based on the actual conversation content; generate real-time speech synthesis parameters based on the adjustment intentions and the initial parameters for outbound speech synthesis; generate real-time outbound speech based on the real-time speech synthesis parameters.

[0188] In one embodiment, when the processor executes the computer program to implement the step of organizing marketing speech knowledge, the processor specifically implements the following steps:

[0189] Organize the graphic and text materials of the graphic and text introduction documents of the products to be marketed, and organize the basic information of the products to be marketed according to the graphic and text materials, and convert the basic information of the products to be marketed into the form of questions and answers for storage; organize the basic information of the marketing promoters, and convert the basic information of the marketing promoters into the form of questions and answers for storage; organize abnormal conversation words.

[0190] In one embodiment, when the processor executes the computer program to implement the steps of obtaining customer information and conversation information of an outbound call conversation and generating initial parameters for outbound speech synthesis based on the customer information and conversation information, the processor specifically implements the following steps:

[0191] Obtain customer information and conversation information of an outbound call conversation; use the GBDT model to predict user connection quality of different speech synthesis parameters based on the customer information and the conversation information to obtain predicted values ​​of the different speech synthesis parameters; and select the speech synthesis parameter with the largest predicted value of the different speech synthesis parameters to obtain the initial parameters for the outbound speech synthesis.

[0192] In one embodiment, when the processor executes the computer program to implement the step of analyzing the adjustment intention in the conversation scenario according to the actual conversation content, the processor specifically implements the following steps:

[0193] Analyze the explicit voice adjustment intention in the actual conversation content; analyze the implicit voice adjustment intention in the actual conversation content; integrate the explicit voice adjustment intention and the implicit voice adjustment intention to form the adjustment intention in the conversation scene.

[0194] In one embodiment, when the processor executes the computer program to implement the step of analyzing the explicit voice adjustment intention in the actual conversation content, the processor specifically implements the following steps:

[0195] TextCNN is used to perform short text classification on the actual conversation content to obtain explicit voice adjustment intentions.

[0196] In one embodiment, when the processor executes the computer program to implement the step of using TextCNN to classify short texts of the actual conversation content to obtain an explicit voice adjustment intention, the processor specifically implements the following steps:

[0197] The speech text corresponding to the actual conversation content is segmented to obtain words; the words are vectorized to obtain word vectors; the word vectors are convolutionally mapped using N convolution kernels to obtain semantic convolution feature information; the semantic convolution feature information is maximum pooled to obtain a pooled value; the pooled value is length-compressed through a fully connected softmax layer to obtain a category probability; and the explicit voice adjustment intention is determined based on the category probability.

[0198] In one embodiment, when the processor executes the computer program to implement the step of analyzing the implicit voice adjustment intention in the actual conversation content, the processor specifically implements the following steps:

[0199] The actual conversation content is classified using a CNN model to obtain implicit voice adjustment intentions.

[0200] In one embodiment, when the processor executes the computer program to implement the step of classifying the actual conversation content using a CNN model to obtain an implicit voice adjustment intention, the processor specifically implements the following steps:

[0201] Segment the speech text within a set context length of the actual conversation content to obtain a number of text segmentations; vectorize the text segmentations in turn to obtain a number of word vectors; perform convolution feature mapping on the word vectors in turn to obtain a number of semantic convolution feature information; calculate the vector difference and vector difference ratio between the semantic convolution feature information in a pairwise manner, and perform convolution feature extraction on the vector difference ratio again to obtain similarity convolution features and connectivity convolution features between the vectors; perform maximum pooling on the similarity convolution features and connectivity convolution features, and perform length compression through a fully connected softmax layer to obtain category probability; determine the implicit speech adjustment intention based on the category probability.

[0202] In one embodiment, when the processor executes the computer program to implement the step of generating the real-time outbound voice according to the real-time voice synthesis parameters, the processor specifically implements the following steps:

[0203] The real-time speech synthesis parameters are sent to a speech synthesis model, so that the speech synthesis model generates real-time outbound speech according to the real-time speech synthesis parameters.

[0204] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0205] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0206] In the several embodiments provided herein, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the various units is merely a logical functional division, and actual implementation may employ other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0207] The steps in the methods of the embodiments of the present invention may be adjusted in order, combined, or deleted as needed. The units in the devices of the embodiments of the present invention may be combined, divided, or deleted as needed. Furthermore, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0208] If this integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, terminal, or network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present invention.

[0209] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for adaptively adjusting speech synthesis parameters, characterized in that: include: Organize marketing skills knowledge; Obtaining customer information and conversation information of an outbound call, and generating initial parameters for outbound speech synthesis based on the customer information and conversation information; Get the actual conversation content; analyzing the adjustment intention in the dialogue scenario based on the actual dialogue content; generating real-time speech synthesis parameters according to the adjustment intention and the initial parameters for outbound speech synthesis; Generate real-time outbound voice according to the real-time voice synthesis parameters; The analyzing the adjustment intention in the conversation scenario according to the actual conversation content includes: analyzing explicit voice adjustment intentions within the actual conversation content; analyzing implicit voice adjustment intentions within the actual conversation content; Integrating the explicit voice adjustment intention and the implicit voice adjustment intention to form an adjustment intention in a dialogue scenario; The analyzing the explicit voice adjustment intention in the actual conversation content includes: Using TextCNN to perform short text classification on the actual conversation content to obtain explicit voice adjustment intentions; The short text classification of the actual conversation content using TextCNN to obtain the explicit voice adjustment intention includes: Segmenting the speech text corresponding to the actual conversation content to obtain words; Vectorizing the words to obtain word vectors; Perform convolution feature mapping on the word vector using N convolution kernels to obtain semantic convolution feature information; Perform maximum pooling on the semantic convolution feature information to obtain the pooling value; The pooled values ​​are passed through a fully connected softmax layer for length compression to obtain the probability of the category; Determine explicit speech adjustment intention based on category probability; The analyzing the implicit voice adjustment intention in the actual conversation content includes: Classifying the actual conversation content using a CNN model to obtain implicit voice adjustment intentions; The classifying the actual conversation content using a CNN model to obtain implicit voice adjustment intentions includes: Segmenting the speech text of the actual conversation content whose context length is within a set range to obtain a plurality of text segmentations; Sequentially vectorizing a plurality of the text segmentations to obtain a plurality of word vectors; Performing convolution feature mapping on a plurality of the word vectors in sequence to obtain a plurality of semantic convolution feature information; Calculating vector differences and vector difference ratios between the plurality of semantic convolution feature information in a pairwise combination manner, and performing convolution feature extraction on the vector difference ratios again to obtain similarity convolution features and connectivity convolution features between the vectors; The similarity convolutional features and the connectivity convolutional features are subjected to maximum pooling and length compression through a fully connected softmax layer to obtain the probability of the category; Determining implicit speech adjustment intentions based on class probabilities.

2. The method for adaptively adjusting speech synthesis parameters according to claim 1, wherein: The said collating marketing speech knowledge includes: Organizing graphic materials of graphic and text introduction documents of products to be marketed, organizing basic information of products to be marketed according to the graphic and text materials, and converting the basic information of products to be marketed into the form of questions and answers for storage; Organize the basic information of marketing promoters and convert it into questions and answers for storage; Organize abnormal conversation words.

3. The method for adaptively adjusting speech synthesis parameters according to claim 1, wherein: The acquiring of customer information and conversation information of the outbound call conversation, and generating initial parameters for outbound speech synthesis according to the customer information and conversation information, includes: Obtain customer information and conversation information for outbound calls; Predicting user connection quality of different speech synthesis parameters using a GBDT model based on the customer information and the conversation information to obtain predicted values ​​of the different speech synthesis parameters; The speech synthesis parameter with the largest predicted value among different speech synthesis parameters is selected to obtain the initial parameter for outbound speech synthesis.

4. The method for adaptively adjusting speech synthesis parameters according to claim 1, wherein: Generating the real-time outbound voice according to the real-time voice synthesis parameter includes: The real-time speech synthesis parameters are sent to a speech synthesis model, so that the speech synthesis model generates real-time outbound speech according to the real-time speech synthesis parameters.

5. A speech synthesis parameter adaptive adjustment device, the device using the speech synthesis parameter adaptive adjustment method according to any one of claims 1 to 4, characterized in that: include: Knowledge organization unit, used to organize marketing knowledge; An initial parameter generation unit, configured to obtain customer information and conversation information of an outbound call, and generate initial parameters for outbound speech synthesis based on the customer information and conversation information; A conversation content acquisition unit, used to acquire actual conversation content; an intention analysis unit, configured to analyze the adjustment intention in the dialogue scenario based on the actual dialogue content; A real-time parameter generating unit, configured to generate real-time speech synthesis parameters according to the adjustment intention and the outbound speech synthesis initial parameters; The speech synthesis unit is used to generate real-time outbound speech according to the real-time speech synthesis parameters.

6. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 4 when executing the computer program.

7. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images