Complex scene-oriented precise voice interaction system

Abstract

The invention relates to the technical field of voice interaction, and discloses a complex scene-oriented precise voice interaction system, which comprises an acoustic feature extraction module for extracting acoustic feature vectors from voice signals and generating acoustic feature representations for suppressing dialect differences; the self-adaptive semantic understanding module is used for generating an initial semantic vector based on the acoustic feature representation, performing semantic enhancement, performing weighted fusion on the historical dialogue semantic vector by combining the dynamically adjusted semantic weighted fusion weight, and generating a user intention representation; the semantic decoding module is used for decoding the user intention representation to generate a precise intention instruction; and the context management module is used for generating a dialogue state vector according to the user intention representation, and generating dialogue context representation by calculating the correlation degree between the dialogue state vector and a historical dialogue semantic vector so as to dynamically adjust the semantic weighted fusion weight. According to the method, the accurate voice recognition rate can be improved, and the context continuity is guaranteed, so that complex and changeable real scene requirements can be better met.

Description

Technical Field

[0001] The present invention relates to the field of voice interaction technology, and in particular to a precise voice interaction system for complex scenarios. Background Art

[0002] With technological advancements, voice interaction systems have been widely used in numerous fields. However, existing voice interaction systems face numerous challenges in complex scenarios. For example, in noisy environments, voice recognition accuracy drops significantly, and background noise cannot be effectively filtered out, leading to command recognition errors. The system's adaptability to different accents, dialects, and specialized terminology is poor, making it difficult to accurately understand user intent. During multi-round conversations, lack of contextual coherence can easily lead to semantic misunderstanding errors. Furthermore, in specific industry applications, the lack of deep integration of specialized domain knowledge makes it impossible to meet complex business needs. This significantly limits the application scope and user experience of voice interaction systems, making them difficult to adapt to the complex and ever-changing requirements of real-world scenarios.

[0003] Therefore, it is urgent to provide a technical solution to solve the above problems. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a precise voice interaction system for complex scenarios.

[0005] In a first aspect, the present invention provides a precise voice interaction system for complex scenarios, the technical solution of which is as follows:

[0006] The acoustic feature extraction module is used to extract acoustic feature vectors from the target speech signal and generate acoustic feature representations that suppress dialect differences through an adversarial training network that includes a dialect classifier;

[0007] An adaptive semantic understanding module is used to generate an initial semantic vector based on the acoustic feature representation, perform semantic enhancement on the initial semantic vector by matching the semantic association matrix of professional terms, and perform weighted fusion on the historical conversation semantic vectors in combination with dynamically adjusted semantic weighted fusion weights to generate a representation of user intent;

[0008] A semantic decoding module, configured to perform mapping decoding on the user intention representation based on a preset instruction template to generate an accurate intention instruction corresponding to the semantic content of the target speech signal;

[0009] A context management module is configured to generate a dialogue state vector based on the user intent representation, calculate the correlation between the dialogue state vector and the historical dialogue semantic vector through an attention mechanism, generate a dynamically updated dialogue context representation, and feed the dialogue context representation back to the adaptive semantic understanding module to dynamically adjust the semantic weighted fusion weight.

[0010] The beneficial effects of the precise voice interaction system for complex scenarios of the present invention are as follows:

[0011] The system of the present invention can solve many problems of existing voice interaction systems in noisy environments, accents and dialects, professional terminology and multi-round dialogue scenarios, improve the accuracy of voice recognition, enhance system adaptability, ensure context coherence, and deeply integrate professional domain knowledge, thereby expanding the scope of application and optimizing user experience, so as to better adapt to the requirements of complex and changing real-world scenarios.

[0012] Based on the above solution, the precise voice interaction system for complex scenarios of the present invention can be further improved as follows.

[0013] In an optional manner, the acoustic feature extraction module is specifically used to:

[0014] Inputting the target speech signal into a deep neural network to extract the acoustic feature vector containing speech content and dialect characteristics;

[0015] Constructing the adversarial training network comprising a generator and a discriminator; the generator is composed of the deep neural network and is used to generate acoustic feature representations that suppress dialect differences; the discriminator is a dialect classifier and is used to distinguish feature representations of different dialect categories;

[0016] Based on the optimization goal, the parameters of the generator and the discriminator are optimized through adversarial training, and the acoustic feature vector is input into the optimized generator to obtain the acoustic feature representation.

[0017] In an optional manner, the adaptive semantic understanding module is specifically used to:

[0018] The acoustic feature representation is input into the semantic encoder to generate the initial semantic vector, which is expressed as:

[0019] v0=Encoder(H inv )

[0020] Where v0 is the initial semantic vector, H inv is the acoustic feature representation, Encoder(·) is the encoder function;

[0021] By calculating the similarity between the initial semantic vector and the professional term semantic association matrix, matching term vectors are screened for semantic enhancement to obtain an enhanced semantic vector, which is expressed as follows:

[0022]

[0023] Where, α irepresents the semantic matching degree between the initial semantic vector and the i-th professional term in the professional term semantic association matrix, v1 represents the enhanced semantic vector, represents the i-th professional term vector in the professional term semantic association matrix, and N represents the number of professional term vectors in the professional term semantic association matrix; Represents the transposed matrix of the initial semantic vector; represents the j-th professional term vector in the professional term semantic association matrix;

[0024] According to the dynamically adjusted semantic weighted fusion weight, the enhanced semantic vector and the historical conversation semantic vector are weightedly fused to generate the user intention representation, which is expressed as follows:

[0025]

[0026] Where, β t represents the tth element of the dynamically adjusted semantic weighted fusion weight β, The key vector representing the tth element in the historical conversation semantic vector, Q current A query vector representing the enhanced semantic vector, represents the tth element of the historical conversation semantic vector, d represents the dimension of the semantic vector, and v intent Represents the user intention representation.

[0027] In an optional manner, the semantic decoding module is specifically configured to:

[0028] Constructing the preset instruction template comprising a plurality of instruction template vectors, each instruction template vector corresponding to an executable operation;

[0029] Calculate the similarity between the user intention representation and each instruction template vector, and calculate the confidence score of each instruction template vector, which is expressed as follows:

[0030]

[0031] Where s j represents the confidence score of the jth instruction template vector, T represents the preset instruction template, T (j) Represents the jth instruction template vector, T (m) represents the mth instruction template vector, M represents the number of instruction template vectors;

[0032] Based on the instruction template vector with the highest confidence score and combined with the semantic details of the user intention representation, the precise intention instruction is generated, which is expressed as follows:

[0033]

[0034] Where, TemplateFill represents the template filling function, which is used to fill v intent The semantic details are injected into the placeholder of the instruction template vector, CMD represents the precise intention instruction.

[0035] In an optional manner, the context management module is specifically configured to:

[0036] Represent the user intention v intent Input a state encoder to generate the dialogue state vector;

[0037] Calculate the attention correlation between the speech state vector and each round vector in the historical dialogue semantic vector, and its expression is:

[0038]

[0039] Where, γ t represents the attention relevance of the t-th round vector in the historical dialogue semantic vector, Represents the dialogue state vector V current The transposed matrix of

[0040] The historical dialogue state vectors are weighted and summed according to the relevance weights to generate the dynamically updated dialogue context representation, which is expressed as follows:

[0041]

[0042] Where c ctx representing the conversation context representation;

[0043] Feedback the dialogue context representation to the adaptive semantic understanding module to dynamically adjust the semantic weighted fusion weight, which is expressed as:

[0044] β=Sigmoid(W β c ctx +b β )

[0045] Where W β represents the weight mapping matrix, b β Represents the bias vector.

[0046] In an optional manner, the method further includes:

[0047] The preprocessing module is used to receive the original speech signal and collaboratively process the original speech signal through the noise suppression model and the speech enhancement model to generate the target speech signal.

[0048] In an optional manner, the professional terminology semantic association matrix is ​​constructed by a professional terminology graph, and the multi-level semantic relationships between terms are mined through a graph neural network.

[0049] In an optional manner, the optimization objective is: the generator minimizes the speech content classification loss while maximizing the dialect classification error of the discriminator, and the discriminator minimizes the dialect classification loss.

[0050] In a second aspect, the present invention provides a precise voice interaction method for complex scenarios, the technical solution of which is as follows:

[0051] Extracting acoustic feature vectors from the target speech signal and generating acoustic feature representations that suppress dialect differences through an adversarial training network that includes a dialect classifier;

[0052] Generating an initial semantic vector based on the acoustic feature representation, semantically enhancing the initial semantic vector by matching the semantic association matrix of professional terms, and weighted fusion of historical conversation semantic vectors using dynamically adjusted semantic weighted fusion weights to generate a representation of user intent;

[0053] Performing mapping decoding on the user intention representation based on a preset instruction template to generate an accurate intention instruction corresponding to the semantic content of the target speech signal;

[0054] A dialogue state vector is generated according to the user intent representation, and the correlation between the dialogue state vector and the historical dialogue semantic vector is calculated through an attention mechanism to generate a dynamically updated dialogue context representation to dynamically adjust the semantic weighted fusion weight.

[0055] The beneficial effects of the present invention's precise voice interaction method for complex scenarios are as follows:

[0056] The method of the present invention can solve many problems of existing voice interaction systems in noisy environments, accents and dialects, professional terminology and multi-round dialogue scenarios, improve the accuracy of voice recognition, enhance system adaptability, ensure context coherence, and deeply integrate professional domain knowledge, thereby expanding the scope of application and optimizing user experience, so as to better adapt to the requirements of complex and changing real-world scenarios.

[0057] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0059] The accompanying drawings are only used to illustrate the embodiments and are not to be considered as limiting the present invention. In addition, the same reference symbols are used to represent the same components throughout the drawings. In the drawings:

[0060] Figure 1 This is a schematic structural diagram of an embodiment of a precise voice interaction system for complex scenarios according to the present invention;

[0061] Figure 2 This is a flow chart of an embodiment of the present invention's method for precise voice interaction in complex scenarios. DETAILED DESCRIPTION

[0062] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0063] Figure 1 The following is a schematic diagram showing the structure of an embodiment of a precise voice interaction system for complex scenarios provided by the present invention. Figure 1 As shown, the system includes:

[0064] The acoustic feature extraction module 110 is used to extract acoustic feature vectors from the target speech signal and generate acoustic feature representations that suppress dialect differences through an adversarial training network including a dialect classifier.

[0065] Among them, the target speech signal refers to the original speech waveform data input by the user through a microphone or other audio acquisition device, and can also be the speech waveform data after noise reduction processing. The acoustic feature vector refers to the high-dimensional representation extracted from the target speech signal, including but not limited to speech content and dialect characteristics. The adversarial training network refers to a joint training architecture including a generator network (feature extractor) and a discriminator network (dialect classifier), which suppresses dialect differences in acoustic features through adversarial games. The acoustic feature representation refers to the dialect invariant features output after processing by the adversarial training network. The dimension is consistent with the acoustic feature vector, but the dialect correlation information is suppressed.

[0066] The adaptive semantic understanding module 120 is used to generate an initial semantic vector based on the acoustic feature representation, semantically enhance the initial semantic vector by matching the semantic association matrix of professional terms, and weightedly fuse the historical conversation semantic vectors in combination with the dynamically adjusted semantic weighted fusion weights to generate a user intent representation.

[0067] The initial semantic vector refers to a d-dimensional vector generated by a semantic encoder performing a nonlinear transformation on the acoustic feature representation, which preliminarily represents the semantic intent in the speech signal. The semantic association matrix for specialized terms is constructed from a specialized term graph, and a graph neural network is used to mine the multi-level semantic relationships between terms. The row vectors of the semantic association matrix for specialized terms represent the d-dimensional semantic vectors of the specialized terms, and the matrix elements reflect the co-occurrence probability or semantic similarity between terms. The dynamically adjusted semantic weighted fusion weights represent the contribution of the historical conversation semantic vectors to the current intent understanding, and the weight values ​​are calculated in real time using an attention mechanism. The historical conversation semantic vector refers to the sequence of semantic vectors stored in the previous k rounds of conversation, with each round corresponding to a d-dimensional vector. The user intent representation is a d-dimensional vector that integrates domain-enhanced semantics and historical context.

[0068] The semantic decoding module 130 is used to perform mapping decoding on the user intention representation based on a preset instruction template to generate an accurate intention instruction corresponding to the semantic content of the target speech signal.

[0069] A preset instruction template is a template containing multiple instruction template vectors, each of which contains fixed text and parameter placeholders. A precise intent instruction is a structured instruction output by the semantic decoding module 130, which includes the operation type, target device, and parameter value. Precise intent instructions are used to drive downstream devices to execute user instructions, ensuring operational accuracy in complex scenarios. For example, if you enter "Turn off all lights after the meeting," the precise intent instruction triggers the conference room lights-off sequence.

[0070] The context management module 140 is used to generate a dialogue state vector based on the user intent representation, calculate the correlation between the dialogue state vector and the historical dialogue semantic vector through the attention mechanism, generate a dynamically updated dialogue context representation, and feed the dialogue context representation back to the adaptive semantic understanding module to dynamically adjust the semantic weighted fusion weight.

[0071] The dialogue state vector is a low-dimensional vector obtained by compressing and encoding the user intent representation. It represents the contextual state of the current dialogue. The dialogue state vector is used to calculate the correlation with historical dialogues and generate a dynamic context representation. The dialogue context representation is a d-dimensional vector formed by weighted fusion of historical dialogue state vectors, dynamically reflecting the semantic context of multiple rounds of dialogue.

[0072] In an optional manner, the acoustic feature extraction module 110 is specifically configured to:

[0073] The target speech signal is input into a deep neural network to extract the acoustic feature vector containing speech content and dialect characteristics.

[0074] Among them, the target speech signal is expressed as T represents the duration of the target speech signal, and F represents the frequency of each frame. Speech content includes phonemes and semantics, and dialect characteristics include pronunciation habits and tone patterns. The acoustic feature vector is expressed as

[0075] Construct the adversarial training network comprising a generator and a discriminator; the generator is composed of the deep neural network and is used to generate acoustic feature representations that suppress dialect differences; the discriminator is a dialect classifier and is used to distinguish feature representations of different dialect categories.

[0076] The generator consists of a 4-layer residual CNN with an output dimension of d=256; the discriminator consists of a 2-layer fully connected network with an output dimension equal to the number of dialect categories.

[0077] Based on the optimization goal, the parameters of the generator and the discriminator are optimized through adversarial training, and the acoustic feature vector is input into the optimized generator to obtain the acoustic feature representation.

[0078] The optimization goal is: the generator minimizes the speech content classification loss while maximizing the dialect classification error of the discriminator, and the discriminator minimizes the dialect classification loss.

[0079] Specifically, the generator loss function is: is the mathematical expectation operator, which represents the statistical average of the input speech signal X; C(·; θ C ) represents the speech content classifier, and the parameter is θ C ;logC(H;θ C ) represents the log-likelihood probability of speech content classification; 1-logD(H; θ D ) represents the logarithmic error of the dialect classifier in discriminating the generated features; λ represents the adversarial loss weight coefficient. The discriminator loss function is: θ D Represents the trainable parameters of the discriminator network.

[0080] In an optional manner, the adaptive semantic understanding module 120 is specifically configured to:

[0081] The acoustic feature representation is input into the semantic encoder to generate the initial semantic vector, which is expressed as:

[0082] v0=Encoder(H inv )

[0083] Where v0 is the initial semantic vector, H inv is the acoustic feature representation, and Encoder(·) is the encoder function.

[0084] By calculating the similarity between the initial semantic vector and the professional term semantic association matrix, matching term vectors are screened for semantic enhancement to obtain an enhanced semantic vector, which is expressed as follows:

[0085]

[0086] Where, α i represents the semantic matching degree between the initial semantic vector and the i-th professional term in the professional term semantic association matrix, v1 represents the enhanced semantic vector, represents the i-th professional term vector in the professional term semantic association matrix, and N represents the number of professional term vectors in the professional term semantic association matrix; Represents the transposed matrix of the initial semantic vector; represents the jth professional term vector in the professional term semantic association matrix.

[0087] According to the dynamically adjusted semantic weighted fusion weight, the enhanced semantic vector and the historical conversation semantic vector are weightedly fused to generate the user intention representation, which is expressed as follows:

[0088]

[0089] Where, β t represents the tth element of the dynamically adjusted semantic weighted fusion weight β, The key vector representing the tth element in the historical conversation semantic vector, Q current A query vector representing the enhanced semantic vector, represents the tth element of the historical conversation semantic vector, d represents the dimension of the semantic vector, and v intent Represents the user intention representation.

[0090] In an optional manner, the semantic decoding module 130 is specifically configured to:

[0091] Constructing the preset instruction template comprising a plurality of instruction template vectors, each instruction template vector corresponding to an executable operation;

[0092] Calculate the similarity between the user intention representation and each instruction template vector, and calculate the confidence score of each instruction template vector, which is expressed as follows:

[0093]

[0094] Where s j represents the confidence score of the jth instruction template vector, T represents the preset instruction template, T (j) Represents the jth instruction template vector, T (m) Represents the mth instruction template vector, and M represents the number of instruction template vectors.

[0095] Based on the instruction template vector with the highest confidence score and combined with the semantic details of the user intention representation, the precise intention instruction is generated, which is expressed as follows:

[0096]

[0097] Where, TemplateFill represents the template filling function, which is used to fill v intent The semantic details are injected into the placeholder of the instruction template vector, CMD represents the precise intention instruction.

[0098] In an optional manner, the context management module 140 is specifically configured to:

[0099] Represent the user intention v intent Input a state encoder to generate the dialogue state vector;

[0100] Calculate the attention correlation between the speech state vector and each round vector in the historical dialogue semantic vector, and its expression is:

[0101]

[0102] Where, γ t represents the attention relevance of the t-th round vector in the historical dialogue semantic vector, Represents the dialogue state vector V current The transposed matrix of

[0103] The historical dialogue state vectors are weighted and summed according to the relevance weights to generate the dynamically updated dialogue context representation, which is expressed as follows:

[0104]

[0105] Where c ctx represents the conversation context representation.

[0106] Feedback the dialogue context representation to the adaptive semantic understanding module to dynamically adjust the semantic weighted fusion weight, which is expressed as:

[0107] β=Sigmoid(W β c ctx +b β )

[0108] Where W β represents the weight mapping matrix, b β Represents the bias vector.

[0109] In an optional manner, the method further includes:

[0110] The preprocessing module is used to receive the original speech signal and collaboratively process the original speech signal through the noise suppression model and the speech enhancement model to generate the target speech signal.

[0111] Specifically, specialized voice acquisition equipment is used to receive the user's raw voice signal. Signal processing algorithms (such as short-time energy calculation and spectral analysis) are then applied to extract representative features from the raw voice signal. The noise suppression model analyzes and judges the input voice features based on knowledge learned during training (for example, a neural network model uses internal parameters such as weights and biases). The raw voice signal is then fed into a pre-trained voice enhancement model, which optimizes the voice features based on its internal structure and parameters. The denoised voice features obtained by the noise suppression model and the optimized voice features obtained by the voice enhancement model are then combined according to a specific fusion strategy. This can be a simple weighted summation approach, where different weights are assigned to the results of the two models based on their importance, or a more complex nonlinear fusion method that fully integrates the characteristics of both models to form a comprehensive optimized voice feature. Using appropriate signal synthesis techniques (such as the inverse short-time Fourier transform), the comprehensively optimized voice features are converted back into a time-domain voice signal, the target voice signal. In this process, it is necessary to ensure that the synthesized speech signal is naturally coherent in the waveform, without obvious distortion or abrupt connection points. Ultimately, the target speech signal is generated, which has lower noise interference, higher voice quality, and better intelligibility and clarity than the original speech signal.

[0112] The technical solution of this embodiment can solve many problems of existing voice interaction systems in noisy environments, accents and dialects, professional terminology and multi-round dialogue scenarios, improve the accuracy of voice recognition, enhance system adaptability, ensure context coherence, and deeply integrate professional domain knowledge, thereby expanding the scope of application and optimizing the user experience, so as to better adapt to the complex and changing real-world scenario requirements.

[0113] Figure 2 The flowchart of an embodiment of a precise voice interaction method for complex scenarios provided by the present invention is shown. Figure 2 As shown, the following steps are included:

[0114] S1. Extract acoustic feature vectors from the target speech signal and generate acoustic feature representations that suppress dialect differences through an adversarial training network that includes a dialect classifier.

[0115] S2. Generate an initial semantic vector based on the acoustic feature representation, perform semantic enhancement on the initial semantic vector by matching the semantic association matrix of professional terms, and perform weighted fusion on the historical conversation semantic vectors in combination with the dynamically adjusted semantic weighted fusion weights to generate a representation of user intent;

[0116] S3, performing mapping decoding on the user intention representation based on a preset instruction template to generate an accurate intention instruction corresponding to the semantic content of the target speech signal;

[0117] S4. Generate a dialogue state vector based on the user intent representation, calculate the correlation between the dialogue state vector and the historical dialogue semantic vector through an attention mechanism, and generate a dynamically updated dialogue context representation to dynamically adjust the semantic weighted fusion weight.

[0118] The technical solution of this embodiment can solve many problems of existing voice interaction systems in noisy environments, accents and dialects, professional terminology and multi-round dialogue scenarios, improve the accuracy of voice recognition, enhance system adaptability, ensure context coherence, and deeply integrate professional domain knowledge, thereby expanding the scope of application and optimizing the user experience, so as to better adapt to the complex and changing real-world scenario requirements.

[0119] Furthermore, the above embodiments provide systems that implement their functions using only the division of the above functional modules as examples. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to actual conditions to complete all or part of the functions described above. Furthermore, the systems and method embodiments provided in the above embodiments share the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.

[0120] The above description is merely a preferred embodiment of the present invention and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present invention is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in the present invention.

[0121] It should be noted that the terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects and to define a specific order or precedence. Where appropriate, the order used for similar objects may be interchanged, such that the embodiments of the present application described herein can be implemented in an order other than the order shown or described.

[0122] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A precise voice interaction system for complex scenarios, characterized by: The system comprises: The acoustic feature extraction module is used to extract acoustic feature vectors from the target speech signal and generate acoustic feature representations that suppress dialect differences through an adversarial training network that includes a dialect classifier; An adaptive semantic understanding module is used to generate an initial semantic vector based on the acoustic feature representation, perform semantic enhancement on the initial semantic vector by matching the semantic association matrix of professional terms, and perform weighted fusion on the historical conversation semantic vectors in combination with dynamically adjusted semantic weighted fusion weights to generate a representation of user intent; A semantic decoding module, configured to perform mapping decoding on the user intention representation based on a preset instruction template to generate an accurate intention instruction corresponding to the semantic content of the target speech signal; A context management module is configured to generate a dialogue state vector based on the user intent representation, calculate the correlation between the dialogue state vector and the historical dialogue semantic vector through an attention mechanism, generate a dynamically updated dialogue context representation, and feed the dialogue context representation back to the adaptive semantic understanding module to dynamically adjust the semantic weighted fusion weight.

2. The precise voice interaction system for complex scenarios according to claim 1, characterized in that: The acoustic feature extraction module is specifically used for: Inputting the target speech signal into a deep neural network to extract the acoustic feature vector containing speech content and dialect characteristics; Constructing the adversarial training network comprising a generator and a discriminator; the generator is composed of the deep neural network and is used to generate acoustic feature representations that suppress dialect differences; the discriminator is a dialect classifier and is used to distinguish feature representations of different dialect categories; Based on the optimization goal, the parameters of the generator and the discriminator are optimized through adversarial training, and the acoustic feature vector is input into the optimized generator to obtain the acoustic feature representation.

3. The precise voice interaction system for complex scenarios according to claim 1, characterized in that: The adaptive semantic understanding module is specifically used to: The acoustic feature representation is input into the semantic encoder to generate the initial semantic vector, which is expressed as: v0=Encoder(H inv ) Where v0 is the initial semantic vector, H inv is the acoustic feature representation, Encoder(·) is the encoder function; By calculating the similarity between the initial semantic vector and the professional term semantic association matrix, matching term vectors are screened for semantic enhancement to obtain an enhanced semantic vector, which is expressed as: Where, α i represents the semantic matching degree between the initial semantic vector and the i-th professional term in the professional term semantic association matrix, v1 represents the enhanced semantic vector, represents the i-th professional term vector in the professional term semantic association matrix, and N represents the number of professional term vectors in the professional term semantic association matrix; Represents the transposed matrix of the initial semantic vector; represents the j-th professional term vector in the professional term semantic association matrix; According to the dynamically adjusted semantic weighted fusion weight, the enhanced semantic vector and the historical conversation semantic vector are weightedly fused to generate the user intention representation, which is expressed as follows: Where, β t represents the tth element of the dynamically adjusted semantic weighted fusion weight β, The key vector representing the tth element in the historical conversation semantic vector, Q current A query vector representing the enhanced semantic vector, represents the tth element of the historical conversation semantic vector, d represents the dimension of the semantic vector, and v intent Represents the user intention representation.

4. The precise voice interaction system for complex scenarios according to claim 3 is characterized in that: The semantic decoding module is specifically used for: Constructing the preset instruction template comprising a plurality of instruction template vectors, each instruction template vector corresponding to an executable operation; Calculate the similarity between the user intention representation and each instruction template vector, and calculate the confidence score of each instruction template vector, which is expressed as follows: Where s j represents the confidence score of the jth instruction template vector, T represents the preset instruction template, T (j) Represents the jth instruction template vector, T (m) represents the mth instruction template vector, M represents the number of instruction template vectors; Based on the instruction template vector with the highest confidence score and combined with the semantic details of the user intention representation, the precise intention instruction is generated, which is expressed as follows: Where, TemplateFill represents the template filling function, which is used to fill v intent The semantic details are injected into the placeholder of the instruction template vector, CMD represents the precise intention instruction.

5. The precise voice interaction system for complex scenarios according to claim 4 is characterized in that: The context management module is specifically used to: Represent the user intention v intent Input a state encoder to generate the dialogue state vector; Calculate the attention correlation between the speech state vector and each round vector in the historical dialogue semantic vector, and its expression is: Where, γ t represents the attention relevance of the t-th round vector in the historical dialogue semantic vector, Represents the dialogue state vector V current The transposed matrix of The historical dialogue state vectors are weighted and summed according to the relevance weights to generate the dynamically updated dialogue context representation, which is expressed as follows: Where c ctx representing the conversation context representation; Feedback the dialogue context representation to the adaptive semantic understanding module to dynamically adjust the semantic weighted fusion weight, which is expressed as: β=Sigmoid(W β ·c ctx +b β ) Where W β represents the weight mapping matrix, b β Represents the bias vector.

6. The precise voice interaction system for complex scenarios according to claim 1, characterized in that: Also includes: The preprocessing module is used to receive the original speech signal and collaboratively process the original speech signal through the noise suppression model and the speech enhancement model to generate the target speech signal.

7. The precise voice interaction system for complex scenarios according to claim 1, characterized in that: The professional term semantic association matrix is ​​constructed by a professional term graph, and the multi-level semantic relationships between terms are mined through a graph neural network.

8. The precise voice interaction system for complex scenarios according to claim 2, characterized in that: The optimization goal is: the generator minimizes the speech content classification loss while maximizing the dialect classification error of the discriminator, and the discriminator minimizes the dialect classification loss.

9. A precise voice interaction method for complex scenarios, characterized by: The method comprises: Extracting acoustic feature vectors from the target speech signal and generating acoustic feature representations that suppress dialect differences through an adversarial training network that includes a dialect classifier; Generating an initial semantic vector based on the acoustic feature representation, semantically enhancing the initial semantic vector by matching the semantic association matrix of professional terms, and weighted fusion of historical conversation semantic vectors using dynamically adjusted semantic weighted fusion weights to generate a representation of user intent; Performing mapping decoding on the user intention representation based on a preset instruction template to generate an accurate intention instruction corresponding to the semantic content of the target speech signal; A dialogue state vector is generated according to the user intent representation, and the correlation between the dialogue state vector and the historical dialogue semantic vector is calculated through an attention mechanism to generate a dynamically updated dialogue context representation to dynamically adjust the semantic weighted fusion weight.

10. The precise voice interaction method for complex scenarios according to claim 9, characterized in that: The step of extracting an acoustic feature vector from the target speech signal and generating an acoustic feature representation that suppresses dialect differences through an adversarial training network including a dialect classifier further includes: Inputting the target speech signal into a deep neural network to extract the acoustic feature vector containing speech content and dialect characteristics; Constructing the adversarial training network comprising a generator and a discriminator; the generator is composed of the deep neural network and is used to generate acoustic feature representations that suppress dialect differences; the discriminator is a dialect classifier and is used to distinguish feature representations of different dialect categories; Based on the optimization goal, the parameters of the generator and the discriminator are optimized through adversarial training, and the acoustic feature vector is input into the optimized generator to obtain the acoustic feature representation.

Images