Text processing method, electronic device, dialogue system and automobile

Abstract

The present invention relates to a text processing method, an electronic device, a dialogue system and a car. The method according to one embodiment of the present invention comprises: embedding a text into a word to obtain a word embedding vector; obtaining a task identifier; inputting the word embedding vector and the task identifier into a trained conversion tag model to obtain a task type and a conversion tag, wherein the conversion tag comprises a first conversion tag representing a part of the text that needs to be converted and a second conversion tag representing a part of the text that does not need to be converted; fusing the task type and the text part corresponding to the first conversion tag to obtain fusion information; inputting the fusion information into a trained text conversion model, and performing a corresponding conversion on the text part to be converted according to the task type to obtain a conversion result. The present invention can achieve simultaneous support for text standardization and inverse text standardization, as well as simultaneous improvement of the accuracy of two tasks, text standardization and inverse text standardization.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing, and in particular to a text processing method, electronic equipment, a dialogue system and a car. Background Art

[0002] The intelligent data cockpit dialogue system in autonomous driving includes modules such as speech recognition, natural language processing and speech synthesis. The text used in these modules often needs to be processed to facilitate modeling and improve readability. Among them, text normalization (TN) and inverse text normalization (ITN) play an important role. The role of text normalization is to realize the conversion of text handwritten format to spoken format, such as converting Arabic numerals in a Chinese sentence into Chinese characters, which is mainly used in the preprocessing process of speech synthesis. Inverse text normalization is the reverse process. After obtaining the recognized text, in order to facilitate user reading, the Chinese character information is converted into a form that is easier for users to read. It is mainly used in the processing stage after speech recognition, such as converting Chinese characters into corresponding digital information through a mapping rule comparison table for the text after speech recognition.

[0003] There are relatively mature solutions for both text normalization and inverse text normalization, but their goal is only to solve one of the two tasks and it is impossible to solve both tasks at the same time, which increases the overall complexity of the dialogue system and the maintenance cost of the system. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a text processing method, an electronic device, a dialogue system, a storage medium and a car to achieve integrated processing of inverse text normalization and text normalization of text information.

[0005] In a first aspect, the present invention provides a text processing method, comprising:

[0006] Embed the text into words to get word embedding vectors;

[0007] Get the task identifier;

[0008] Inputting the word embedding vector and the task identifier into a trained conversion label model to obtain a task type and a conversion label, wherein the conversion label includes a first conversion label representing a part of the text that needs to be converted and a second conversion label representing a part of the text that does not need to be converted;

[0009] Fusing the task type and the text portion corresponding to the first conversion mark to obtain fusion information;

[0010] The fusion information is input into a trained text conversion model, and the text portion to be converted is converted accordingly according to the task type to obtain a conversion result.

[0011] In a specific implementation, obtaining the task identifier includes:

[0012] Acquire a scenario in which the text is to be applied, wherein the scenario to be applied includes a first scenario corresponding to the need for text normalization and a second scenario corresponding to the need for inverse text normalization;

[0013] The task identifier is determined according to a scenario to be applied, wherein the task identifier includes a first identifier corresponding to a first scenario and a second identifier corresponding to a second scenario.

[0014] In a specific implementation, the step of inputting the word embedding vector and the task identifier into a trained conversion labeling model to obtain a task type and a conversion label includes:

[0015] Inputting the word embedding vector and the task identifier into a trained conversion tag model to obtain a tag sequence, wherein the tag sequence includes a task type bit and a conversion tag bit;

[0016] The task type at the task type position includes a text standardization task corresponding to the first identifier and an inverse text standardization task corresponding to the second identifier;

[0017] The conversion mark on the conversion mark bit includes the first conversion mark and the second conversion mark.

[0018] In a specific implementation, the step of fusing the task type and the text portion corresponding to the first conversion mark to obtain fusion information includes:

[0019] The task type, the text portion corresponding to the first conversion marker, and the context of the text corresponding to the first conversion marker are fused to obtain fusion information.

[0020] The fusing of the task type, the text portion corresponding to the first conversion marker, and the context of the text corresponding to the first conversion marker includes:

[0021] The task type, the text portion corresponding to the first conversion marker, and the context of the text corresponding to the first conversion marker are concatenated.

[0022] In a specific implementation, the step of inputting the fusion information into a trained text conversion model and converting the to-be-converted text portion accordingly according to the task type to obtain a conversion result includes:

[0023] If the task type is a text standardization task, the text conversion model performs text standardization processing on the part of the text to be converted to obtain a text standardization conversion result;

[0024] If the task type is an inverse text standardization task, the text conversion model performs inverse text standardization processing on the part of the text to be converted to obtain an inverse text standardization conversion result.

[0025] In a specific embodiment, the method further comprises:

[0026] The conversion result is concatenated with the part of the text that does not need to be converted to obtain a complete text.

[0027] In a specific implementation, the trained conversion tagging model is composed of M layers of bidirectional LSTM networks; the trained text conversion model is composed of N Transformer networks, where M and N are natural numbers.

[0028] In a specific embodiment, the method further includes the step of training a Norm network composed of the conversion tag model and the text conversion model.

[0029] In a specific embodiment,

[0030] The loss function used is Conduct training, where:

[0031]

[0032] The output conditional probability of the Norm network is:

[0033]

[0034] in,

[0035] Norm(.) represents the output conditional probability, sl represents the lth element of the sequence, θ represents the parameters of the network, and D represents the number of samples input in each batch during training.

[0036] In a second aspect, the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores a program, and when the program is executed by the processor, the method according to the first aspect is implemented.

[0037] In a third aspect, the present invention provides a storage medium, wherein the memory stores a program, and when the program is executed by the processor, the method according to the first aspect is implemented.

[0038] In a fourth aspect, the present invention provides an in-vehicle dialogue system, comprising the electronic device according to the second aspect and an input-output device.

[0039] In a fifth aspect, the present invention provides a car, comprising the in-vehicle dialogue system according to the fourth aspect.

[0040] The above one or more technical solutions of the present invention have at least one or more of the following

[0041] Beneficial effects:

[0042] The present invention introduces a task identifier when processing text, obtains the task type and the text part that needs to be converted according to the task type through a conversion tag model, and integrates them as the input of a text conversion model. The text conversion model performs corresponding text conversion according to the task type, and can simultaneously support text standardization and inverse text standardization. Not only can it simplify the complexity of the overall intelligent dialogue system, but also because the same text conversion model is used, the accuracy of the two tasks of text standardization and inverse text standardization can be simultaneously improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the drawings represent similar components, wherein:

[0044] Figure 1 is a flowchart of a text processing method according to an embodiment of the present invention;

[0045] Figure 2 is a schematic diagram of a conversion markup model structure according to an embodiment of the present invention;

[0046] Figure 3 is a schematic diagram of a text conversion model structure according to an embodiment of the present invention;

[0047] Figure 4 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0049] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as program codes, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Non-temporary computer-readable storage media include any suitable media that can store program codes, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, and the like.

[0050] See attached Figure 1 , Figure 1 1 is a flowchart of a text processing method according to an embodiment of the present invention, comprising the following steps:

[0051] S10: embed the text into words to obtain word embedding vectors.

[0052] The text is word-embedded to obtain an embedding vector X=(x1, ..., x L ), where L represents the sentence length. In a specific embodiment, for example, the text obtained from the speech recognition module of the intelligent cockpit dialogue system is "Dial 138 xxxxxxxx", then L = 15. Further, assuming that the embedding dimension n is 100, each xi (1≤i≤L) in the embedding vector is 100-dimensional, and finally a 17*100-dimensional matrix is ​​formed. In another specific embodiment, for example, the text obtained from the speech recognition module of the intelligent cockpit dialogue system is "This journey is about 215.6km", then L = 12.

[0053] Of course, those skilled in the art will appreciate that the text used in the present invention is not limited to being obtained from the smart cockpit dialogue system, and this is just an example.

[0054] S20: Obtain a task identifier.

[0055] In a specific embodiment, for example, a task identifier T of 0 represents a text standardization task, and a task identifier T of 1 represents an inverse text standardization task.

[0056] Those skilled in the art will appreciate that, in an alternative solution, 0 may be used to represent an inverse text normalization task, and 1 may be used to represent a text normalization task, and the present invention does not limit this.

[0057] In a specific embodiment, step S20 includes the following steps:

[0058] Acquire a scenario in which the text is to be applied, wherein the scenario to be applied includes a first scenario corresponding to the need for text normalization and a second scenario corresponding to the need for inverse text normalization;

[0059] Specifically, if the text processing using the method of the present invention is used in a post-speech recognition processing scenario, such as for converting text and presenting it on a human-computer interaction interface for intelligent driving, then the scenario corresponds to inverse text standardization. For example, the text "Dial 138 xxxxxxxx" in the above example is converted to "Dial 138********" and presented on the human-computer interaction interface for user reading convenience. For another example, if the text processing using the method of the present invention is used in a speech synthesis pre-processing scenario, that is, converting text and then broadcasting it by voice, then the scenario corresponds to text standardization. For example, the text "This journey is about 215.6 km" in the above example is converted to "This journey is about two hundred and fifteen point six kilometers" to facilitate conversion to voice for broadcasting.

[0060] The task identifier is determined according to a scenario to be applied, wherein the task identifier includes a first identifier corresponding to a first scenario and a second identifier corresponding to a second scenario.

[0061] In the two examples above, the task identifier corresponding to the first scenario is 1, and the task identifier corresponding to the second scenario is 0.

[0062] S30: Input the word embedding vector and the task identifier into a trained conversion tag model to obtain a task type task and a conversion tag sig, wherein the conversion tag includes a first conversion tag representing a part of the text that needs to be converted and a second conversion tag representing a part of the text that does not need to be converted.

[0063] In a specific embodiment, for ease of calculation, the task identifier is converted into a vector of the same dimension as each word in the word embedding vector, for example, 100 dimensions in the above example.

[0064] Dimensionally align the task identifiers and embedding vectors X = (x1, ..., x L ) to form a matrix and input it into the trained conversion labeling model. For example, in the above example of the text "Dial phone 138 xxxxxxxx", the embedding vector X = (x1, ..., x17) and the task identifier T = 1 form a matrix and input it into the trained conversion labeling model.

[0065] In a specific embodiment, the conversion tag model adopts the LSTM model commonly used in the field of natural language processing. The LSTM model is particularly suitable for processing categorical data whose input sequence is a time problem. More preferably, the present invention adopts Figure 2The bidirectional LSTM model shown, the model of the bidirectional LSTM trained on the input sequence is two LSTMs instead of one LSTM. The first in the input sequence is the original sample, and the second is the reverse sample of the input sequence. As can be seen from the figure, the forward layer and the reverse layer are connected to the output layer together, which contains 6 shared weights w1-w6, which can provide additional context for the network and can learn the problem faster and more comprehensively. More preferably, in order to learn the problem more comprehensively, the conversion labeling model of the present invention can adopt an M-layer (e.g., 2-layer) bidirectional LSTM network, and the output layer of the bidirectional LSTM network of the first layer is connected to the input of the bidirectional LSTM network of the second layer, and so on.

[0066] In a specific embodiment, the conversion tag model outputs a tag sequence, wherein the tag sequence includes a task type bit and a conversion tag bit. The task type in the task type bit includes a text standardization task corresponding to the first identifier and a reverse text standardization task corresponding to the second identifier; and the conversion tag in the conversion tag bit includes the first conversion tag and the second conversion tag.

[0067] For example, for the example of the text "Dial 138 xxxxxxxx" and T=1, the tag sequence is obtained: "ITNSSSSBEEEEEEEEEE", where the beginning of the tag sequence is the task type bit, which is the task type ITN corresponding to T=1 in this example. However, those skilled in the art will understand that whether the task type bit is positioned at the beginning or other positions of the tag sequence can be set as needed, and the present invention does not limit this. For the ITN task, the text portion that needs to be converted in the text is "138 xxxxxxxx", and the conversion tag uses B as the start tag, E as the end tag, and the middle part is represented by E. However, those skilled in the art will understand that it is possible to represent the conversion tag with other letters or forms, and the present invention does not limit this. On the other hand, for the ITN task, the text portion that does not need to be converted in the text is "Dial a phone call", and the conversion tag is represented by S. However, those skilled in the art will understand that it is possible to represent the conversion tag with other letters or forms, and the present invention does not limit this.

[0068] In another example, for example, for the example of the text "This journey is approximately 215.6 km" and T=0, a tag sequence is obtained: "TNSSSSSBEEEEEE", wherein the beginning of the tag sequence is the task type bit, which in this example is the task type TN corresponding to T=0. However, those skilled in the art will appreciate that whether the task type bit is positioned at the beginning or other positions of the tag sequence can be set as needed, and the present invention does not limit this. For the TN task, the text portion that needs to be converted in the text is "215.6 km", and the conversion mark uses B as the start mark and E as the end mark, and the middle part is represented by E. However, those skilled in the art will appreciate that it is possible to represent the conversion mark with other letters or forms, and the present invention does not limit this. On the other hand, for the TN task, the text portion that does not need to be converted in the text is "This journey is approximately", and the conversion mark is represented by S. However, those skilled in the art will appreciate that it is possible to represent the conversion mark with other letters or forms, and the present invention does not limit this.

[0069] Those skilled in the art will appreciate that, for text standardization tasks and inverse text standardization tasks, which parts of the text need to be converted can be pre-set as rules, and the method of the present invention performs identification according to the rules.

[0070] S40: Fusing the task type and the text portion corresponding to the first conversion mark to obtain fusion information.

[0071] According to the first conversion mark, the corresponding text part is extracted and merged with the task type. For example, for the example of the text "Dial 138 xxxxxxxx" and T=1, the vector corresponding to the text part "138 xxxxxxxx" to be converted and the vector corresponding to ITN need to be merged (e.g., concatenated).

[0072] Preferably, in order to achieve a better conversion effect, the present invention also splices the task type, the text part to be converted, and the context of the text part to be converted. In a specific example, for example, the n words before and after the text part to be converted are fused as contexts, and n can be set as needed, such as 2. For example, for the text "Dial 138 xxxxxxxx", the two preceding words of the text part "138 xxxxxxxx" to be converted are "telephone", and there is no following context, then only the preceding words are taken as the context in this example.

[0073] S50: Inputting the fusion information into a trained text conversion model, and performing corresponding conversion on the text portion to be converted according to the task type to obtain a conversion result.

[0074] If the task type is a text standardization task, the text conversion model performs text standardization processing on the part of the text to be converted to obtain a text standardization conversion result.

[0075] If the task type is an inverse text standardization task, the text conversion model performs inverse text standardization processing on the part of the text to be converted to obtain an inverse text standardization conversion result.

[0076] For example, for the example of "Call 138 xxxxxxxx" and T=1, the output after conversion is "138********". For another example, for the example of "This journey is about 215.6km" and T=0, the output after conversion is "two hundred and fifteen point six kilometers".

[0077] In a specific embodiment, the text conversion model of the present invention adopts a Transformer network, such as Figure 3 As shown, the Transformer network consists of an encoder-decoder, and each block contains an attention mechanism. More preferably, in order to improve the effect of the conversion, multiple Transformer networks can be used, and the output of the previous network is used as the input of the subsequent network.

[0078] In a specific embodiment, for final application, the conversion result is finally concatenated with the part of the text that does not need to be converted to obtain a complete text, such as "call 138********", "this journey is approximately 215.6 kilometers".

[0079] Those skilled in the art will appreciate that in order to apply the above conversion mark model and text conversion model, training is required. In a specific embodiment of the present invention, the conversion mark model and the text conversion model constitute a Norm network, which is trained using training samples.

[0080] In a specific embodiment, the training uses the following loss function: Conduct training,

[0081]

[0082] The output conditional probability of the Norm network is:

[0083]

[0084] in,

[0085] Norm(.) represents the output conditional probability, sl represents the lth element of the sequence, θ represents the parameters of the network, and D represents the number of samples input in each batch during training.

[0086] For example, the total training samples are 1,000 texts, the number of input samples in each batch is 100, and the model is trained by inputting them 10 times.

[0087] Specifically, a back propagation method may be used, and the training is terminated when the loss function reaches a predetermined threshold. Afterwards, the test sample is used for testing.

[0088] The present invention introduces a task identifier when processing text, obtains the task type and the text part that needs to be converted according to the task type through a conversion tag model, and integrates them as the input of a text conversion model. The text conversion model performs corresponding text conversion according to the task type, and can simultaneously support text standardization and inverse text standardization. Not only can it simplify the complexity of the overall intelligent dialogue system, but also because the same text conversion model is used, the accuracy of the two tasks of text standardization and inverse text standardization can be simultaneously improved.

[0089] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0090] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a server, or installed from a storage device. When the computer program is executed, the above-mentioned functions defined in the method of the embodiment of the present invention are implemented.

[0091] The text processing method according to an embodiment of the present invention may be applied to the dialogue system of the intelligent data cockpit described above, for example, as a function of a natural language processing module.

[0092] To this end, the present invention also provides an electronic device, such as Figure 4 As shown, a processor (such as a central processing unit, a graphics processing unit, etc.) and a storage medium may be included. A program is stored in the storage medium, and when the program is executed by the processor, a text processing method according to an embodiment of the present invention is implemented.

[0093] The program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form.

[0094] Furthermore, when the program is executed by the processor, it can also realize speech recognition and speech synthesis functions, that is, the electronic device includes software function modules such as a speech recognition module, a natural language processing module and a speech synthesis module.

[0095] To this end, the present invention also provides a vehicle-mounted dialogue system, comprising the above electronic device and an input / output device, wherein the processor and the storage medium can be connected to each other and to the input / output device via a bus.

[0096] In a specific embodiment, the input / output device of the vehicle-mounted dialogue system of the present invention includes a touch screen, a microphone, a speaker, etc., which are used to constitute the hardware and software required for the intelligent data cockpit dialogue system in autonomous driving.

[0097] The present invention further provides a computer-readable storage medium, in which a program is stored. When the program is executed by a processor, a text processing method according to an embodiment of the present invention is implemented.

[0098] It should be noted that the computer-readable storage medium of the present invention can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0099] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0100] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A text processing method, characterized in that: include: Embed the text into words to get word embedding vectors; Get the task identifier; Inputting the word embedding vector and the task identifier into a trained conversion label model to obtain a task type and a conversion label, wherein the conversion label includes a first conversion label representing a part of the text that needs to be converted and a second conversion label representing a part of the text that does not need to be converted; Fusing the task type and the text portion corresponding to the first conversion mark to obtain fusion information; Inputting the fusion information into a trained text conversion model, and performing corresponding conversion on the text part to be converted according to the task type to obtain a conversion result; The obtaining of the task identifier comprises: Acquire a scenario in which the text is to be applied, wherein the scenario to be applied includes a first scenario corresponding to the need for text normalization and a second scenario corresponding to the need for inverse text normalization; The task identifier is determined according to a scenario to be applied, wherein the task identifier includes a first identifier corresponding to a first scenario and a second identifier corresponding to a second scenario.

2. The method according to claim 1, characterized in that The step of inputting the word embedding vector and the task identifier into a trained conversion label model to obtain a task type and a conversion label includes: Inputting the word embedding vector and the task identifier into a trained conversion tag model to obtain a tag sequence, wherein the tag sequence includes a task type bit and a conversion tag bit; The task type at the task type position includes a text standardization task corresponding to the first identifier and an inverse text standardization task corresponding to the second identifier; and The conversion mark on the conversion mark bit includes the first conversion mark and the second conversion mark.

3. The method according to claim 1, characterized in that The step of fusing the task type and the text portion corresponding to the first conversion mark to obtain fusion information includes: The task type, the text portion corresponding to the first conversion marker, and the context of the text corresponding to the first conversion marker are fused to obtain fusion information.

4. The method according to claim 3, characterized in that The fusing of the task type, the text portion corresponding to the first conversion marker, and the context of the text corresponding to the first conversion marker includes: The task type, the text portion corresponding to the first conversion marker, and the context of the text corresponding to the first conversion marker are concatenated.

5. The method according to claim 1, characterized in that The step of inputting the fusion information into a trained text conversion model and converting the text part to be converted according to the task type to obtain a conversion result includes: If the task type is a text standardization task, the text conversion model performs text standardization processing on the part of the text to be converted to obtain a text standardization conversion result; If the task type is an inverse text standardization task, the text conversion model performs inverse text standardization processing on the part of the text to be converted to obtain an inverse text standardization conversion result.

6. The method according to claim 5, characterized in that The method further comprises: The conversion result is concatenated with the part of the text that does not need to be converted to obtain a complete text.

7. The method according to claim 1, characterized in that The trained conversion labeling model is composed of an M-layer bidirectional LSTM network; The trained text conversion model is composed of N Transformer networks, where M and N are natural numbers.

8. The method according to any one of claims 1 to 7, characterized in that The method also includes the step of training the Norm network composed of the conversion tag model and the text conversion model.

9. The method according to claim 8, characterized in that Using loss function Conduct training, where: The output conditional probability of the Norm network is: in, Norm(.) represents the output conditional probability, sl represents the lth element of the sequence, θ represents the parameters of the network, and D represents the number of samples input in each batch during training.

10. An electronic device comprising a processor and a memory, characterized in that: The memory stores a program, and when the program is executed by the processor, the method according to any one of claims 1 to 9 is implemented.

11. A storage medium, characterized in that: The storage medium stores a program, and when the program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

12. A vehicle-mounted dialogue system, characterized in that: include: The electronic device according to claim 10; Input-output device.

13. A car, characterized in that: Comprising the in-vehicle dialogue system according to claim 12.

Images