Voice processing method, electronic device, storage medium, program product and system

Abstract

This application discloses a speech processing method, electronic device, storage medium, program product, and speech processing system. The method includes acquiring target speech; recognizing the target speech and obtaining a first response result and a confidence level of the first response result based on the recognition result; in response to a confidence level of the first response result being less than a preset threshold, sending at least one of the target speech and corresponding feature information, along with historical processing information, to a server for processing to obtain a second response result for the target speech; wherein the historical processing information includes historical recognition results and historical response results corresponding to historical speech; and responding to the target speech according to the second response result. Through the above methods, this application can improve the accuracy of speech response.

Description

Technical Field

[0001] This application relates to the field of speech processing, and in particular to a speech processing method, electronic device, storage medium, program product, and system. Background Technology

[0002] In current voice interaction systems, large language models are a crucial component for language understanding and generation. However, the reasoning speed of large language models is relatively slow, leading to excessively long response times in voice interaction. To address this issue, it's common practice to place large language models in the cloud to improve their reasoning efficiency, or to simultaneously utilize both local and cloud processing to accelerate voice interaction. When local processing detects an unreasonable response to the target speech, it delegates the target speech to the cloud for further processing. However, in continuous interaction scenarios, if cloud intervention is needed midway, relying solely on the target speech for response processing may result in errors in speech intent recognition due to a lack of historical information, leading to incorrect responses. Summary of the Invention

[0003] The main purpose of this application is to provide a speech processing method, electronic device, storage medium, program product, and system that can improve the accuracy of speech response.

[0004] To address the aforementioned technical problems, the first technical solution adopted in this application is: providing a speech processing method. This method is applied to a device, which is communicatively connected to a server. The method includes acquiring target speech; recognizing the target speech and obtaining a first response result and a confidence level of the first response result based on the recognition result; in response to a confidence level of the first response result being less than a preset threshold, sending at least one of the target speech and corresponding feature information, along with historical processing information, to the server for processing to obtain a second response result for the target speech; wherein the historical processing information includes historical recognition results and historical response results corresponding to historical speech; and responding to the target speech according to the second response result.

[0005] To address the aforementioned technical problems, the second technical solution adopted in this application is: providing a speech processing method. This method is applied on a server side, which communicates with a device side. The method includes receiving at least one of target speech and corresponding feature information from the device side, along with historical processing information; wherein the historical processing information includes historical recognition results and historical response results corresponding to historical speech from the device side during speech processing; processing the target speech, at least one of the corresponding feature information, and the historical processing information to obtain a second response result for the target speech; and sending the second response result to the device side so that the device side responds to the target speech based on the second response result.

[0006] To address the aforementioned technical problems, the third technical solution adopted in this application is to provide an electronic device. This electronic device includes a memory and a processor. The memory stores program data, which can be executed by the processor to implement the method described in the first or second technical solution.

[0007] To address the aforementioned technical problems, the fourth technical solution adopted in this application is to provide a computer-readable storage medium. This computer-readable storage medium stores program data and can be executed by a processor to implement the methods described in the first and / or second technical solutions.

[0008] To address the aforementioned technical problems, the fifth technical solution adopted in this application is to provide a computer program product. This computer program product stores program data and can be executed by a processor to implement the methods described in the first and / or second technical solutions.

[0009] To solve the above-mentioned technical problems, the sixth technical solution adopted in this application is: to provide a voice processing system, which includes a device end and a server end, and the device end and the server end are communicatively connected to implement the method described in the first and / or second technical solutions.

[0010] The beneficial effects of this application are as follows: After acquiring the target speech, the target speech is processed on the device side to obtain the device side's recognition result of the target speech, the first response result in response to the recognition result, and the confidence level of the first response result. When the confidence level of the first response result is less than a preset threshold, it indicates that the device side's local processing system may not respond accurately to the target speech. Therefore, the target speech and / or the feature information corresponding to the target speech are sent to the server side for a second response processing. In order to avoid response errors caused by the lack of historical information in the server side's processing system, the device side also sends historical processing information to the server side as auxiliary information for the server side's processing of the target speech. Combining historical processing information can make the server side's response to the target speech more accurate. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0012] Figure 1 This is a schematic diagram of a speech processing flow;

[0013] Figure 2This is a flowchart illustrating the first embodiment of the speech processing method of this application;

[0014] Figure 3 This is a flowchart illustrating the second embodiment of the speech processing method of this application;

[0015] Figure 4 This is a flowchart illustrating the third embodiment of the speech processing method of this application;

[0016] Figure 5 This is a flowchart illustrating the fourth embodiment of the speech processing method of this application;

[0017] Figure 6 This is a flowchart illustrating a specific embodiment of the speech processing method of this application;

[0018] Figure 7 This is a schematic diagram of the structure of an embodiment of the electronic device of this application;

[0019] Figure 8 This is a schematic diagram of the structure of an embodiment of the computer-readable storage medium of this application;

[0020] Figure 9 This is a schematic diagram of the structure of an embodiment of the computer program product of this application;

[0021] Figure 10 This is a schematic diagram of the structure of an embodiment of the speech processing system of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0023] The terms "first," "second," etc., used in this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0024] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] In typical voice interaction scenarios, the larger the language model used, the longer its response time. Common large language models generally exhibit noticeable delays in their voice responses. However, in voice interaction, if the response interval exceeds a certain time, such as 0.2 seconds, the dialogue becomes unnatural. Therefore, to minimize response time in voice interaction, one can accelerate the inference speed of the large language model, utilize classifiers and traditional NLP pipelines for traffic splitting, or set up an offline processing system on the local device to process the voice response. When the offline processing system cannot meet the processing needs, a cloud-based voice processing system can then be used. Figure 1 As shown, Figure 1 This is a schematic diagram of a speech processing flow.

[0026] In order to address the issue of potential response errors caused by the lack of historical information in the cloud during continuous interactive scenarios when cloud intervention is used midway, this application proposes the following embodiments.

[0027] Reference Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the speech processing method of this application. The method is applied to a device, which is communicatively connected to a server. The device in this method is equipped with a local speech processing system. This local speech processing system may include an ASR processing module, an NLP processing module, a decision module, and a caching module. Speech is processed by the ASR and NLP processing modules to obtain a response result and a corresponding confidence level. The decision module is used to determine the confidence level of the response result to decide whether the current speech requires server-side intervention. The caching module is used to cache data that needs to be uploaded to the server for processing. The data stored in the caching module may include speech data and / or processed speech feature information data. This method includes, but is not limited to, the following steps.

[0028] S11: Obtain the target speech.

[0029] The target speech is the sound collected by the device's sound recording system that needs to be processed. The device often determines the target speech through a voice wake-up word. When the device recognizes the voice wake-up word, all speech collected within a preset time period thereafter will be considered as sound that needs to be processed and requires a response.

[0030] S12: Recognize the target speech and obtain the first response result and the confidence level of the first response result based on the recognition result.

[0031] S13: In response to the confidence level of the first response result being less than a preset threshold, at least one of the target speech and the corresponding feature information and historical processing information are sent to the server for processing to obtain the second response result of the target speech.

[0032] When the confidence level of the first response is less than a preset threshold, it indicates that the device-side speech processing system may not cover the target speech, and its response to the target speech may be inaccurate. Therefore, server-side intervention is required because the server-side speech processing system is networked, and its speech coverage is larger than that of the device-side offline speech processing system. The data sent to the server can be the target speech, feature information processed from the target speech, or both. Since the feature information is obtained by extracting features from the target speech, its data volume is smaller than that of the target speech data. Therefore, sending only the feature information can reduce the amount of data transmitted between the device and the server, reduce data transmission time, and thus reduce the speech response time.

[0033] Furthermore, to avoid response errors caused by the lack of historical information in the server-side processing system, the device also sends historical processing information to the server as auxiliary information for the server to process the target speech, thereby assisting the server in processing the target speech. This historical processing information includes historical recognition results and historical response results corresponding to historical speech.

[0034] S14: Respond to the target speech according to the second response result.

[0035] After processing the target audio on the server side and obtaining the second response, the server sends the second response to the device. The device then generates the corresponding execution command based on the second response. Alternatively, the server generates the corresponding execution command based on the second response and then sends the execution command to the device.

[0036] Responding to the target voice according to the second response result includes executing the corresponding operation according to the execution instruction corresponding to the second response result, and can also provide a voice reply according to the second response result.

[0037] In this embodiment, after acquiring the target speech, the target speech is processed on the device side to obtain the device's recognition result of the target speech, the first response result in response to the recognition result, and the confidence level of the first response result. When the confidence level of the first response result is less than a preset threshold, it indicates that the device's local processing system may not be accurately responding to the target speech. Therefore, the target speech and / or the feature information corresponding to the target speech are sent to the server for a second response processing. To avoid response errors caused by the server's processing system lacking historical information, the device also sends historical processing information to the server as auxiliary information for the server's processing of the target speech. Combining historical processing information makes the server's response to the target speech more accurate.

[0038] In one embodiment, the voice processing flow on the device side responds to a voice wake-up word. The historical processing information includes the historical recognition results and historical response results from the current time period to the response to the previous voice wake-up word. All voice processing processes from the previous wake-up word to the current time are considered as a complete scenario, and the processing information in this process is used as the historical processing information of the current processing flow.

[0039] In one embodiment, the historical processing information includes the historical recognition results and historical response results of all speech processing processes within a preset number of times or a preset time period prior to the target speech processing process. Processing information from a period prior to the current speech processing process or from at least one speech processing process is used as the historical processing information for the current speech processing process.

[0040] In one embodiment, the two can be combined, and the historical processing information can include the historical recognition results and historical response results of all voice processing processes within a preset number of times or a preset time period before the target voice processing process, from the current time to the response to the previous voice wake-up word.

[0041] In one embodiment, in response to a confidence level greater than or equal to a preset threshold for the first response result, the target voice is responded to according to the first response result. After receiving the first response result on the device side, a corresponding execution instruction is generated based on the first response result. Responding to the target voice according to the first response result includes performing the corresponding operation according to the execution instruction corresponding to the first response result. Further, it may also include providing a voice reply according to the first response result.

[0042] Reference Figure 3 , Figure 3This is a flowchart illustrating a second embodiment of the speech processing method of this application. The method is applied to a server, which communicates with a device. The server in this method is equipped with a server speech processing system. This server speech processing system may include a speech recognition module and a large language model module. Further, it may also include a classification module and a natural language processing module. After the speech is processed by the speech recognition module, it is then processed by the large language model module or the natural language processing module to obtain a response result. The classification module is used to determine whether the speech has passed through the large language model module or the natural language processing module. This method includes, but is not limited to, the following steps.

[0043] S21: Receive at least one of the target speech and the corresponding feature information from the device, and historical processing information.

[0044] The target speech is the sound collected by the device's sound recording system that needs to be processed. The device often determines the target speech through a voice wake-up word. When the device recognizes the voice wake-up word, all speech collected within a preset time period thereafter will be considered as sound that needs to be processed and requires a response.

[0045] The server can also receive feature information corresponding to the target speech. This feature information is obtained by extracting features from the target speech, and its data volume is smaller than that of the target speech itself. When the device only uploads the feature information corresponding to the target speech, compared to uploading the speech data, the amount of data transmitted from the device to the server can be reduced, thereby further reducing the response time of voice interaction.

[0046] For example, a speech recognition module can include an ASR processing module. Taking F-bank features, which ASR can accept, as an example, a 1-second speech sample with a sampling rate of 16kHz has a data size of 16000*32 bits. Using a common setting of 20ms frame shift and 50ms per frame, and employing 80-dimensional F-bank features, the 1-second data is compressed to 50 frames * 80 * 32 bits. If FP16 quantization is used, the data size is further compressed to 1 / 8 of the original. Alternatively, discrete speech tokens can be used. Taking Encoder as an example, using approximately 8 codebooks of 1024 bits each achieves excellent compression efficiency. Therefore, the encoder using Encoder only requires 50 (frames) * 8 * 10 (encoding 1024 tokens) bits.

[0047] The historical processing information includes the historical recognition results and historical response results of the device during the voice processing process.

[0048] In one embodiment, the voice processing flow on the device side responds to a voice wake-up word. The historical processing information includes the historical recognition results and historical response results from the current time period to the response to the previous voice wake-up word. All voice processing processes from the previous wake-up word to the current time are considered as a complete scenario, and the processing information in this process is used as the historical processing information of the current processing flow.

[0049] In one embodiment, the historical processing information includes the historical recognition results and historical response results of all speech processing processes within a preset number of times or a preset time period prior to the target speech processing process. Processing information from a period prior to the current speech processing process or from at least one speech processing process is used as the historical processing information for the current speech processing process.

[0050] In one embodiment, the two can be combined, and the historical processing information can include the historical recognition results and historical response results of all voice processing processes within a preset number of times or a preset time period before the target voice processing process, from the current time to the response to the previous voice wake-up word.

[0051] S22: Process at least one of the target speech and the corresponding feature information and historical processing information to obtain a second response result of the target speech.

[0052] S23: Send a second response result to the device so that the device responds to the target speech based on the second response result.

[0053] The server obtains a second response result based on the received target speech and / or its corresponding feature information, and further combines this with historical processing information. The generated second response result is then sent to the device. Alternatively, the server generates a corresponding execution instruction based on the second response result and sends both the second response result and the corresponding execution instruction to the device.

[0054] Reference Figure 4 , Figure 4 This is a flowchart illustrating the third embodiment of the speech processing method of this application. This method is a further extension of step S22, and includes, but is not limited to, the following steps.

[0055] S31: Input at least one of the target speech and the corresponding feature information into the speech recognition module to obtain the recognition result.

[0056] S32: Input the recognition results and historical processing information into the large language model module to obtain the second response result of the target speech.

[0057] When the server side includes a speech recognition module (which may include an ASR processing module) and a large language model module, the target speech or the feature information corresponding to the target speech can be processed sequentially through the speech recognition module and the large language model module to obtain the corresponding response result.

[0058] The speech recognition model processes the target speech or its corresponding feature information to obtain the recognition result. If the data uploaded to the server includes both feature information and speech data, the recognition result is obtained through the feature information. It can also be further verified based on the recognition result obtained from the target speech to arrive at the final output recognition result. The large language model receives the recognition result input from the speech recognition module and processes it using historical processing information as prompt information to obtain a second response result.

[0059] Reference Figure 5 , Figure 5 This is a flowchart illustrating the fourth embodiment of the speech processing method of this application. This method is a further extension of the above embodiments and includes, but is not limited to, the following steps.

[0060] S41: Input the recognition results into the classification module.

[0061] When the server-side further includes a classification module and a natural language processing module (which may include an NLP processing module), after obtaining the recognition result, it is input into the classification module for discrimination to determine the intent type of the recognition result. Based on the intent type, it determines whether to input the result into the natural language processing module or the large language model module. The classification module and the natural language processing module are used to offload speech processing, thereby reducing the amount of data processed in the large language model module and reducing response time.

[0062] S42: In response to the recognition result being of a single intent, the recognition result is input to the natural speech processing module to obtain the second response result of the target speech by combining historical processing information; in response to the recognition result being of a multi-intent, the recognition result is input to the large language model module to obtain the second response result of the target speech by combining historical processing information.

[0063] If the response result is a single intent, it is processed by the natural language processing module; if it is a multiple intent, it is processed by the large language model module in conjunction with historical processing information.

[0064] If processed by the natural language processing module, the recognition result and historical processing information are input into the natural language processing module to obtain the second response result. The historical processing information will serve as contextual information for the recognition result, assisting the natural language processing module in processing it.

[0065] The following specific embodiment will be used to illustrate the speech processing method of this application in more detail. (Refer to...) Figure 6 , Figure 6 This is a flowchart illustrating a specific embodiment of the speech processing method of this application.

[0066] It can be seen that the local loop on the device side will dominate most of the voice interaction, and local voice data will only be uploaded when the decision-maker determines that cloud intervention is required. The cloud voice processing system only acts as a secondary voice processing system, and uploads cached data only when the local decision-maker determines that secondary intervention is required.

[0067] If the confidence level of the current response result is low, the current target speech or corresponding feature information is cached for uploading to the cloud for secondary processing. The device also needs to transmit the local ASR processing result (the recognition result mentioned above in this application), the NLP result (the response result mentioned above in this application), and even the local instruction execution result (the execution instruction generated based on the response result mentioned above in this application) to the cloud. The cloud then combines this information. For traditional NLP, this forms contextual information (not shown in this route); for large model links, it forms prompt information.

[0068] like Figure 7 As shown, Figure 7 This is a schematic diagram of the structure of an embodiment of the electronic device of this application.

[0069] The electronic device includes a processor 110 and a memory 120.

[0070] Processor 110 controls the operation of electronic devices. Processor 110 may also be referred to as a CPU (Central Processing Unit). Processor 110 may be an integrated circuit chip with signal sequence processing capabilities. Processor 110 may also be a general-purpose processor, a digital signal sequence processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0071] The memory 120 stores the instructions and program data required for the processor 110 to operate. The memory 120 may include a medium capable of storing program instructions, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. Alternatively, it may be a server storing the program instructions, which can send the stored program instructions to other devices for execution or execute the stored program instructions itself.

[0072] The processor 110 is used to respond to the results to implement the methods provided by any embodiment and possible combination of the speech processing method embodiments applied to the device side of this application, or to implement the methods provided by any embodiment and possible combination of the speech processing method embodiments applied to the server side of this application.

[0073] like Figure 8 As shown, Figure 8 This is a schematic diagram of the structure of an embodiment of the computer-readable storage medium of this application.

[0074] One embodiment of the readable storage medium of this application includes a memory 210 that stores program data, which, when executed, implements the methods provided in any embodiment and possible combinations of the speech processing methods of this application.

[0075] The memory 210 may include a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or other media that can store program instructions. Alternatively, it may be a server that stores the program instructions, which can send the stored program instructions to other devices for execution or execute the stored program instructions itself.

[0076] like Figure 9 As shown, Figure 9 This is a schematic diagram of the structure of an embodiment of the computer program product of this application.

[0077] One embodiment of the computer program product of this application includes a memory 310, which stores program data that, when executed, implements the methods provided in any embodiment and possible combinations of the speech processing methods of this application.

[0078] The memory 310 may include a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or other media that can store program instructions. Alternatively, it may be a server that stores the program instructions, which can send the stored program instructions to other devices for execution or execute the stored program instructions itself.

[0079] like Figure 10 As shown, Figure 10 This is a schematic diagram of the structure of an embodiment of the voice processing system of this application.

[0080] One embodiment of the voice processing system of this application includes a device 410 and a server 420. The device 410 may include the electronic device implementing the device-side method in the electronic device embodiments of this application, and the server 420 may include the electronic device implementing the server-side method in the electronic device embodiments of this application. The device 410 and server 420 are communicatively connected to implement the methods provided by any embodiment and possible combinations thereof in the foregoing voice processing method embodiments of this application.

[0081] In summary, after acquiring the target speech, this application processes the target speech on the device side to obtain the device's recognition result, the first response result in response to the recognition result, and the confidence level of the first response result. When the confidence level of the first response result is less than a preset threshold, it indicates that the device's local processing system may not respond accurately to the target speech. Therefore, the target speech and / or the corresponding feature information are sent to the server for a second response processing. To avoid response errors caused by the server's lack of historical information, the device also sends historical processing information to the server as auxiliary information for the server's processing of the target speech. Combining historical processing information makes the server's response to the target speech more accurate.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0083] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0084] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0085] If the integrated units in the other embodiments described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, 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 several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0086] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A speech processing method, characterized in that, Applied to a device that communicates with a server, the method includes: Acquire the target speech; The target speech is identified, and a first response result and the confidence level of the first response result are obtained based on the identification result; If the confidence level of the first response result is less than a preset threshold, the feature information and historical processing information corresponding to the target speech are sent to the server for processing to obtain a second response result of the target speech; wherein, the historical processing information includes historical recognition results and historical response results corresponding to historical speech, and the data volume of the feature information corresponding to the target speech is less than the data volume of the target speech; Respond to the target speech according to the second response result; The server-side component includes a speech recognition module and a large language model module. The step of sending the feature information and historical processing information corresponding to the target speech to the server-side for processing to obtain a second response result of the target speech includes: The feature information corresponding to the target speech is input into the speech recognition module to obtain the recognition result; The recognition result and the historical processing information are input into the large language model module to obtain the second response result of the target speech; The server-side component further includes a classification module and a natural language processing module. After obtaining the recognition result, it includes: The recognition result is input into the classification module; In response to the recognition result being of a single intent, the recognition result is input to the natural language processing module to obtain the second response result of the target speech in combination with the historical processing information; in response to the recognition result being of a multi-intent, the recognition result is input to the large language model module to obtain the second response result of the target speech in combination with the historical processing information. The voice processing flow on the device side is initiated in response to a voice wake-up word, and the historical processing information includes the historical recognition results and the historical response results within the time period from the current time to the response to the previous voice wake-up word.

2. The method according to claim 1, characterized in that, The historical processing information includes the historical recognition results and historical response results of all speech processing processes within a preset number of times or a preset time period prior to the target speech processing process.

3. The method according to claim 1, characterized in that, The method further includes: If the confidence level of the first response result is greater than or equal to the preset threshold, the target speech is responded to according to the first response result.

4. A speech processing method, characterized in that, Applied to the server side, wherein the server side communicates with the device side, the method includes: The device receives feature information and historical processing information corresponding to the target speech from the device. The historical processing information includes historical recognition results and historical response results corresponding to the historical speech during the speech processing process. The amount of data for the feature information corresponding to the target speech is less than the amount of data for the target speech. The feature information and historical processing information corresponding to the target speech are processed to obtain a second response result of the target speech; Send the second response result to the device so that the device responds to the target speech based on the second response result; The server-side includes a speech recognition module and a large language model module. The process of processing the feature information and historical processing information corresponding to the target speech to obtain a second response result of the target speech includes: The feature information corresponding to the target speech is input into the speech recognition module to obtain the recognition result; The recognition result and the historical processing information are input into the large language model module to obtain the second response result of the target speech; The server-side component further includes a classification module and a natural language processing module. After obtaining the recognition result, it includes: The recognition result is input into the classification module; In response to the recognition result being of a single intent, the recognition result is input to the natural language processing module to obtain the second response result of the target speech in combination with the historical processing information; in response to the recognition result being of a multi-intent, the recognition result is input to the large language model module to obtain the second response result of the target speech in combination with the historical processing information. The voice processing flow on the device side is initiated in response to a voice wake-up word, and the historical processing information includes the historical recognition results and the historical response results within the time period from the current time to the response to the previous voice wake-up word.

5. An electronic device, characterized in that, It includes a memory and a processor, the memory being used to store program data, the program data being executable by the processor to implement the method as described in any one of claims 1-3 or claim 4.

6. A computer-readable storage medium, characterized in that, It stores program data and can be executed by a processor to implement the method as described in any one of claims 1-4.

7. A computer program product, characterized in that, It stores program data and can be executed by a processor to implement the method as described in any one of claims 1-4.

8. A voice processing system, characterized in that, It includes a device and a server, wherein the device and the server are connected in communication to implement the method as described in any one of claims 1-4.

Images