Heterogeneous speech recognition engine load balancing method, device, medium and electronic device

By determining the load allocation ratio based on the recognition accuracy of the voice recognition engine and allocating voice recognition requests, the problem of only balancing the number of requests rather than load balancing in the prior art is solved, and the actual load balancing of the voice recognition engine is realized.

CN115083418BActive Publication Date: 2024-11-12CHINA TELECOM CORP LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110268112.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-03-12
Publication Date
2024-11-12
Estimated Expiration
2041-03-12

AI Technical Summary

Technical Problem

The existing voice recognition system does not consider the recognition capabilities of each voice recognition engine when allocating voice recognition requests, resulting in only balancing the number of requests, but not load balancing in the actual sense.

Method used

By receiving voice recognition requests from different regions, the load allocation ratio of the online voice recognition engine in each region is obtained according to the area identification. The ratio is determined based on the recognition accuracy of the offline voice recognition engine, and the voice recognition request is allocated according to the ratio to achieve the Load balancing of each voice recognition engine.

Benefits of technology

It realizes the allocation of voice recognition requests based on the recognition capabilities of each voice recognition engine, ensuring load balancing in the actual sense and avoiding the defects caused by balancing the number of requests.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115083418B_ABST
    Figure CN115083418B_ABST
Patent Text Reader

Abstract

The present invention provides a method for load balancing of heterogeneous speech recognition engines, comprising: receiving speech recognition requests from different regions; obtaining the load distribution ratio of online speech recognition engines in each region according to region identifiers in the speech recognition requests of the different regions, wherein online speech recognition engines of different architectures are deployed in each region, and the load distribution ratio of online speech recognition engines in each region is determined based on the recognition accuracy of offline speech recognition engines of different architectures in the region during optimization; and distributing speech recognition requests from the different regions according to the load distribution ratio of online speech recognition engines in each region, thereby distributing speech recognition requests from different regions according to the speech recognition capabilities of each engine, thereby achieving load balancing of each speech recognition engine in a practical sense. The present invention also provides a device, medium, and electronic device for load balancing of heterogeneous speech recognition engines.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of speech recognition, and in particular to a method, device, medium and electronic device for load balancing of heterogeneous speech recognition engines. Background Art

[0002] Speech recognition technology is a technology that enables machines to understand human language through the process of recognition and understanding. Speech recognition technology is a key technology for human-computer interaction in information technology. It has been widely used in call centers, value-added services, and enterprise information systems. With the in-depth application of speech recognition in new application fields such as voice search and voice control, speech recognition technology has been praised by industry authorities as a key technology that may trigger a revolution in the field of human-computer interface.

[0003] At present, speech recognition engines from different manufacturers can be deployed in the speech recognition system, and each engine processes requests from different sources respectively. For example, after receiving a speech recognition request, a domain name polling method is usually used, that is, multiple records are set for the domain name and polling is performed to achieve load balancing between the various speech recognition engines. Specifically, polling is performed by domain name polling with the same load distribution ratio to achieve load balancing between the various speech recognition engines. However, when implementing the inventive concept of the present invention, the inventors found that there are the following technical problems in the related art: when allocating speech recognition requests, the recognition capabilities of each speech recognition engine are not taken into account. Therefore, the related art only achieves the balance of the number of speech recognition requests, but does not achieve the load balancing of each speech recognition engine in a practical sense.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method, device, medium and electronic device for load balancing of heterogeneous speech recognition engines, thereby at least to a certain extent avoiding the technical defects of the related art that only the balance of the number of speech recognition requests is achieved without achieving load balancing of each speech recognition engine in a practical sense.

[0006] Other features and advantages of the present invention will become apparent from the following detailed description, or may be learned in part by practice of the present invention.

[0007] According to a first aspect of an embodiment of the present invention, a method for load balancing of heterogeneous speech recognition engines is provided, comprising: receiving speech recognition requests from different regions; obtaining the load distribution ratio of online speech recognition engines in each region according to region identifiers in the speech recognition requests from the different regions, wherein online speech recognition engines of different architectures are deployed in each region, and the load distribution ratio of online speech recognition engines in each region is determined based on the recognition accuracy of offline speech recognition engines of different architectures in the region during an optimization process; and distributing speech recognition requests from the different regions according to the load distribution ratio of online speech recognition engines in each region.

[0008] In some embodiments of the present invention, the method further includes: obtaining training samples from a recording platform in each area, wherein the training samples include recording files for training an offline speech recognition engine and recognition results of the recording files; optimizing offline speech recognition engines of different architectures in each area using the recording files for training the offline speech recognition engine and the recognition results of the recording files to obtain an optimized offline speech recognition engine.

[0009] In some embodiments of the present invention, the method also includes: obtaining a test sample from a recording platform in each area, the test sample including a recording file for testing the optimized offline speech recognition engine and a recognition result of the recording file; using the recording file for testing the optimized offline speech recognition engine and the recognition result of the recording file, testing the optimized offline speech recognition engine of different architectures in each area to obtain the recognition accuracy of the optimized offline speech recognition engine of different architectures in each area.

[0010] In some embodiments of the present invention, the method further includes: determining the load distribution ratio of the online speech recognition engine in each region based on the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization of different architectures in each region; and storing the load distribution ratio of the online speech recognition engine in each region in a designated database.

[0011] In some embodiments of the present invention, determining the load distribution ratio of the online speech recognition engine in each area based on the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization of different architectures in each area includes: calculating the specific proportion of the online speech recognition engine when responsible for allocating traffic based on the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization; determining the load distribution ratio of the online speech recognition engine in each area based on the specific proportions of multiple online speech recognition engines in each area when responsible for allocating traffic.

[0012] In some embodiments of the present invention, the method further includes: receiving a first update task, wherein the first update task includes an identifier of an online speech recognition engine to be updated; and updating the online speech recognition engine in each region according to the identifier of the online speech recognition engine to be updated and the optimized offline speech recognition engine.

[0013] In some embodiments of the present invention, the method further includes: receiving a second update task, wherein the second update task includes the current load distribution ratio of the online speech recognition engine in each area; and updating the historical load distribution ratio of the online speech recognition engine in each area of ​​the designated database using the current load distribution ratio of the online speech recognition engine in each area.

[0014] According to a second aspect of an embodiment of the present invention, a device for load balancing of heterogeneous speech recognition engines is provided, comprising: a first receiving module, used to receive speech recognition requests from different regions; a first acquisition module, used to obtain the load distribution ratio of the online speech recognition engine in each region according to the region identifier in the speech recognition requests from the different regions, wherein online speech recognition engines of different architectures are deployed in each region, and the load distribution ratio of the online speech recognition engines in each region is determined based on the recognition accuracy of offline speech recognition engines of different architectures in the region during an optimization process; and an allocation module, used to allocate the speech recognition requests from the different regions according to the load distribution ratio of the online speech recognition engines in each region.

[0015] In some embodiments of the present invention, the device method also includes: a second acquisition module, used to obtain training samples from the recording platform of each area, the training samples including recording files for training offline speech recognition engines and recognition results of the recording files; a training module, using the recording files for training offline speech recognition engines and the recognition results of the recording files, to optimize offline speech recognition engines of different architectures in each area to obtain optimized offline speech recognition engines.

[0016] In some embodiments of the present invention, the device also includes: a third acquisition module, used to obtain test samples from the recording platform of each area, the test samples including recording files for testing the optimized offline speech recognition engine and the recognition results of the recording files; a testing module, using the recording files for testing the optimized offline speech recognition engine and the recognition results of the recording files, to test the optimized offline speech recognition engines of different architectures in each area, and obtain the recognition accuracy of the optimized offline speech recognition engines of different architectures in each area.

[0017] In some embodiments of the present invention, the device also includes: a distribution ratio determination module, which is used to determine the load distribution ratio of the online speech recognition engine in each area based on the recognition accuracy of the offline speech recognition engine after optimization of different architectures in each area and the recognition accuracy of the offline speech recognition engine before optimization; a storage module, which is used to store the load distribution ratio of the online speech recognition engine in each area in a designated database.

[0018] In some embodiments of the present invention, the above-mentioned distribution ratio determination module includes: a distribution ratio determination module, which is used to calculate the specific proportion of the online speech recognition engine when being responsible for allocating traffic based on the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization; a distribution ratio determination sub-module, which is used to determine the load distribution ratio of the online speech recognition engine in each area based on the specific proportion of multiple online speech recognition engines in each area when being responsible for allocating traffic.

[0019] In some embodiments of the present invention, the device also includes: a second receiving module, used to receive a first update task, wherein the first update task includes an identifier of an online speech recognition engine to be updated; and a first update module, used to update the online speech recognition engine in each area according to the identifier of the online speech recognition engine to be updated and the optimized offline speech recognition engine.

[0020] In some embodiments of the present invention, the device also includes: a third receiving module, used to receive a second update task, wherein the second update task includes the current load distribution ratio of the online speech recognition engine in each area; a second update module, using the current load distribution ratio of the online speech recognition engine in each area, to update the historical load distribution ratio of the online speech recognition engine in each area of ​​the specified database.

[0021] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method for load balancing of heterogeneous speech recognition engines as described in the first aspect of the above embodiment.

[0022] According to a fourth aspect of an embodiment of the present invention, a computer-readable medium is provided, on which a computer program is stored. When the program is executed by a processor, the method for load balancing of heterogeneous speech recognition engines as described in the first aspect of the above embodiment is implemented.

[0023] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0024] In the technical solution provided by some embodiments of the present invention, speech recognition requests from different regions are received, and the load distribution ratio of the online speech recognition engine in each region is obtained according to the region identifier in the speech recognition requests from different regions, wherein online speech recognition engines with different architectures are deployed in each region, and the load distribution ratio of the online speech recognition engine in each region is determined based on the recognition accuracy of offline speech recognition engines with different architectures in the region during the optimization process, and then the speech recognition requests from different regions are distributed according to the load distribution ratio of the online speech recognition engine in each region. In this way, the speech recognition requests from different regions are distributed according to the speech recognition capabilities of each engine, thereby achieving load balancing of each speech recognition engine in a practical sense.

[0025] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings herein are incorporated into and constitute a part of the specification, showing embodiments consistent with the present invention, and together with the specification, are used to explain the principles of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0027] Figure 1 A schematic diagram of a system for load balancing of heterogeneous speech recognition engines according to an embodiment of the present invention is schematically shown;

[0028] Figure 2 Schematically shows a flow chart of a method for load balancing of heterogeneous speech recognition engines according to an embodiment of the present invention;

[0029] Figure 3 Schematically shows a flow chart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0030] Figure 4 Schematically shows a flow chart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0031] Figure 5 Schematically shows a flow chart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0032] Figure 6 Schematically shows a flow chart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0033] Figure 7Schematically shows a flow chart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0034] Figure 8 Schematically shows a flow chart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0035] Fig. 9 A block diagram schematically shows a device for load balancing of heterogeneous speech recognition engines according to an embodiment of the present invention;

[0036] Fig.10 A block diagram schematically shows a device for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0037] Fig.11 A block diagram schematically shows a device for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0038] Fig.12 A block diagram schematically shows a device for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0039] Fig.13 A block diagram schematically shows a device for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0040] Fig.14 A block diagram schematically shows a device for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0041] Fig.15 A block diagram schematically shows a device for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention;

[0042] Fig.16 A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0043] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present invention will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.

[0044] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, those skilled in the art will appreciate that the technical solution of the present invention can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

[0045] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0046] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0047] Figure 1 The following is a schematic diagram of a system for load balancing of heterogeneous speech recognition engines according to an embodiment of the present invention.

[0048] like Figure 1 As shown, the system 100 for load balancing of heterogeneous speech recognition engines includes a distribution unit 110, a main control unit 120, an optimization unit 130, a training effect evaluation unit 140, and online speech recognition engines of different architectures and offline speech recognition engines of different architectures. Among them, online speech recognition engines of different architectures are deployed in an online recognition environment to perform recognition processing on online speech recognition requests of different regions. For example, the online recognition environment includes regions 1 to N, and online speech recognition engines of different architectures can be deployed in each region of regions 1 to N. Specifically, speech recognition engine A1 (abbreviated as ASR engine A1), speech recognition engine B1 (abbreviated as ASR engine B1), etc. can be deployed in region 1. Speech recognition engine AN1 (abbreviated as ASR engine AN1), speech recognition engine BN1 (abbreviated as ASR engine BN1), etc. can be deployed in region N.

[0049] Among them, offline speech recognition engines of different architectures are deployed in an offline optimization environment to optimize offline speech recognition requests in different regions to improve the recognition accuracy of each offline speech recognition engine. For example, the offline optimization environment includes regions 1 to N, and offline speech recognition engines of different architectures can be deployed in each region of regions 1 to N. Specifically, speech recognition engine A (abbreviated as ASR engine A), speech recognition engine B (abbreviated as ASR engine B1), etc. can be deployed in each region.

[0050] In some embodiments of the present invention, the responsible allocation unit 110 is used to receive speech recognition requests (abbreviated as ASR requests) from different regions, and send each speech recognition request to each region according to the region identifier in the speech recognition request. Then, the load distribution ratio of the online speech recognition engine in each region is obtained according to the region identifier in the speech recognition request, and the load distribution ratio is determined based on the recognition accuracy of each offline speech recognition engine in the training effect evaluation unit 140. And according to the load distribution ratio of each region, the speech recognition requests sent to each region are distributed, so that the speech recognition requests can be distributed according to the recognition ability of each online speech recognition engine, that is, the load balancing of each speech recognition engine in a practical sense is realized, and the defects brought by only realizing the balancing of the number of speech recognition requests in the related art are effectively avoided. For example, according to the load distribution ratio of the online speech recognition engine in region 1, the speech recognition requests in region 1 are distributed. The load distribution ratio of the online speech recognition engine in region 1 can be determined according to the recognition accuracy of each offline speech recognition engine in region 1 in the training effect evaluation unit 140.

[0051] In some embodiments of the present invention, the optimization unit 130 obtains training samples from the recording platform in each area, and the training samples include recording files for training offline speech recognition engines and recognition results of the recording files. The training samples of each area are used to optimize multiple offline speech recognition engines in each area, so that the recognition accuracy of the offline speech recognition engines in each area can be improved, and the optimized offline speech recognition engines are obtained. For example, the optimization unit 130 obtains the training samples of the area from the recording platform of area 1, and uses the training samples to optimize the offline speech recognition engines of different architectures in area 1 to improve the recognition accuracy of the offline speech recognition engines of different architectures in area 1, and obtains the optimized offline speech recognition engines of different architectures in area 1. For another example, the optimization unit 130 obtains the training samples of the area from the recording platform of area N, and uses the training samples to optimize the offline speech recognition engines of different architectures in area N to improve the recognition accuracy of the offline speech recognition engines of different architectures in area N, and obtains the optimized offline speech recognition engines of different architectures in area N.

[0052] In some embodiments of the present invention, after the optimization of the offline speech recognition engine in each area is completed by the optimization unit 130, the training effect evaluation unit 140 obtains a test sample from the recording platform in each area, and the test sample includes a recording file for testing the optimized offline speech recognition engine and the recognition result of the recording file. The test samples of each area are used to test the optimized offline speech recognition engines of different architectures in each area to obtain the recognition accuracy of the optimized offline speech recognition engines of different architectures in each area. For example, the training effect evaluation unit 140 obtains the test sample of the area from the recording platform of area 1, and the test sample is used to test the optimized offline speech recognition engines of different architectures in area 1 to obtain the recognition accuracy of the optimized offline speech recognition engines of different architectures in area 1. For another example, the training effect evaluation unit 140 obtains the test sample of the area from the recording platform of area N, and the test sample is used to test the optimized offline speech recognition engines of different architectures in area N to obtain the recognition accuracy of the optimized offline speech recognition engines of different architectures in area N. In this embodiment, the training effect evaluation unit 140 can avoid the problem of difficulty in automation and standardization caused by the inconsistency in the output result format of speech recognition engines with different architectures, because in this example, the recording file and the speech recognition result file are in a unified format. In this way, the training effect evaluation unit 140 can automatically perform real-time testing on the optimized offline speech recognition engines with different architectures in each region.

[0053] In some embodiments of the present invention, the main control unit 120 controls the optimization unit 120 to optimize the offline speech recognition engines of different architectures in each area, and controls the training effect evaluation unit 140 to test the optimized offline speech recognition engines of different architectures in each area, and calculates the load distribution ratio of the online speech recognition engine in the area according to the recognition accuracy of the optimized offline speech recognition engines of different architectures in each area, and then sends the load distribution ratio of the corresponding online speech recognition engine to each area in the online recognition environment, so as to distribute the speech recognition requests from different areas according to the load distribution ratio of the online speech recognition engine.

[0054] In some embodiments of the present invention, the main control unit 120 may also control the responsible allocation unit 110 to allocate speech recognition requests from different areas.

[0055] In some embodiments of the present invention, the above-mentioned main control unit 120 can also use the optimized offline speech recognition engine trained by the optimization unit 130 to regularly update the online speech recognition engines in each area of ​​the online recognition environment. And the current online speech recognition engine load distribution ratio tested by the training effect evaluation unit 140 is used to regularly update the historical online speech recognition engine load distribution ratio of each area in the online recognition environment. This helps to dynamically adjust the business load of online speech recognition engines of different architectures in each area of ​​the online recognition environment.

[0056] In some embodiments of the present invention, the distribution unit 110, the main control unit 120, the optimization unit 130, and the training effect evaluation unit 140 can be implemented using the springcloud microservice architecture. In this embodiment, the components of each unit are independently deployed daemons, and all provide Restful API business access interfaces for the main control unit to call. In addition, the front-end web interface of the main control unit 120 is implemented using vue.js. The load distribution unit 110 can be implemented based on the istio component to achieve dynamic traffic scheduling and distributed call tracking. The optimization unit 130 can be implemented based on the open source Talend API Tester. The training effect evaluation unit 140 can be implemented based on the automatic scoring tool for encapsulating the NIST SCLITE component. For example, the adaptation function of the speech recognition engine business interface and the training interface can be implemented based on the open source Talend API Tester, and the automatic scoring tool is implemented for encapsulating the NIST SCLITE component.

[0057] Figure 2 The flowchart of the method for load balancing of heterogeneous speech recognition engines according to an embodiment of the present invention is schematically shown.

[0058] like Figure 2 As shown, the method for load balancing of heterogeneous speech recognition engines may include steps S210 to S230.

[0059] In step S210, speech recognition requests from different regions are received.

[0060] In step S220, based on the region identifier in the speech recognition request of the different regions, the load distribution ratio of the online speech recognition engine in each region is obtained, wherein online speech recognition engines with different architectures are deployed in each region, and the load distribution ratio of the online speech recognition engine in each region is determined based on the recognition accuracy of offline speech recognition engines with different architectures in the region during the optimization process.

[0061] In step S230, the speech recognition requests in the different regions are distributed according to the load distribution ratio of the online speech recognition engines in each region.

[0062] The method can receive speech recognition requests from different regions, and obtain the load distribution ratio of online speech recognition engines in each region according to region identifiers in the speech recognition requests from different regions, wherein online speech recognition engines with different architectures are deployed in each region, and the load distribution ratio of online speech recognition engines in each region is determined based on the recognition accuracy of offline speech recognition engines with different architectures in the region during the optimization process, and then the speech recognition requests from different regions are distributed according to the load distribution ratio of online speech recognition engines in each region. In this way, the speech recognition requests from different regions are distributed according to the speech recognition capabilities of each engine, thereby achieving load balancing of each speech recognition engine in a practical sense.

[0063] In one embodiment of the present invention, the speech recognition requests in different regions may be speech recognition requests in different locations, for example, speech recognition requests in Beijing, speech recognition requests in Shanghai, speech recognition requests in Hubei, speech recognition requests in Heilongjiang, etc., but not limited thereto.

[0064] In one embodiment of the present invention, the speech recognition requests for different regions include identifiers of different regions. For example, the speech recognition request for Beijing includes BJ, the speech recognition request for Shanghai includes SH, the speech recognition request for Hubei includes HB, the speech recognition request for Heilongjiang includes HLJ, etc., but the present invention is not limited thereto.

[0065] In one embodiment of the present invention, the load distribution ratio of the online speech recognition engine in each region can be obtained according to the region identifier in the speech recognition request of different regions. For example, the load distribution ratio of the online speech recognition engine in Beijing can be obtained according to the above BJ. In this embodiment, the above regions can refer to different regional computer rooms, for example, Beijing computer room, Shanghai computer room, Hubei computer room, Heilongjiang computer room. Among them, servers are deployed in each computer room, and speech recognition engines of different architectures are deployed on servers in each computer room, that is, online speech recognition engines of different architectures are deployed in each region.

[0066] In one embodiment of the present invention, the load distribution ratio of the online speech recognition engine in each region is determined based on the recognition accuracy of the offline speech recognition engines of different architectures in the region during the optimization process. Figure 1 , the load distribution ratio of the online speech recognition engine is calculated according to the recognition accuracy of the offline speech recognition engine after optimization of different architectures in each area of ​​the training effect evaluation unit 140

[0067] In one embodiment of the present invention, the speech recognition requests in different regions are allocated according to the load distribution ratio of the online speech recognition engines in each region. For example, the speech recognition requests in different regions are allocated according to the load distribution ratio determined by the recognition accuracy of the optimized offline speech recognition engines of different architectures, so that the speech recognition requests can be allocated to each online speech recognition engine in a targeted manner, thus achieving load balancing of the speech recognition engines in a true sense.

[0068] In one embodiment of the present invention, the above-mentioned method for load balancing of heterogeneous speech recognition engines can be applied to an intelligent voice client system. The intelligent voice client system can be applied to various application scenarios, such as finance, insurance, logistics, operators, etc.

[0069] Figure 3 A flowchart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0070] like Figure 3 As shown, the above method may further include steps S310 and S320.

[0071] In step S310, a training sample is obtained from the recording platform of each area, wherein the training sample includes a recording file for training an offline speech recognition engine and a recognition result of the recording file.

[0072] In step S320, the offline speech recognition engine of different architectures in each region is optimized using the recording file for training the offline speech recognition engine and the recognition result of the recording file to obtain an optimized offline speech recognition engine.

[0073] The method can utilize the recording files used to train the offline speech recognition engine and the recognition results of the recording files to optimize the offline speech recognition engines of different architectures in each region to obtain an optimized offline speech recognition engine, thereby improving the recognition capability of the offline speech recognition engine and making the recognition accuracy of the optimized offline speech recognition engine higher.

[0074] In one embodiment of the present invention, the training samples are obtained from the recording platform of each region. For example, the training samples are obtained from the recording platform in Beijing, the training samples are obtained from the recording platform in Shanghai, the training samples are obtained from the recording platform in Hubei, the training samples are obtained from the recording platform in Heilongjiang, and so on, but not limited to this. In this embodiment, the recording files in the training samples can be audio files that are difficult to recognize, that is, when the recording files are recognized by the online speech recognition engine, the recognition results obtained are very different from the meaning expressed by the real user. The recognition results of the recording files in the above training samples are manually marked recognition results. Among them, the recording files can be recording files of various business scenarios.

[0075] refer to Figure 1 , by using the recording file used to train the offline speech recognition engine and the recognition result of the recording file, the offline speech recognition engines of different architectures in regions 1 to N are optimized by the optimization unit 130 to obtain optimized offline speech recognition engines, which can improve the recognition accuracy of offline speech recognition engines of different architectures. Specifically, the optimization unit 130 is used to adapt the interface of the offline speech recognition engine to achieve the training optimization of the offline recognition engine.

[0076] Figure 4 A flowchart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0077] like Figure 4 As shown, the above method may further include step S410 and step S420.

[0078] In step S410, a test sample is obtained from the recording platform of each region, wherein the test sample includes a recording file for testing the optimized offline speech recognition engine and a recognition result of the recording file.

[0079] In step S420, the optimized offline speech recognition engine of different architectures in each region is tested using the recording file used to test the optimized offline speech recognition engine and the recognition result of the recording file to obtain the recognition accuracy of the optimized offline speech recognition engine of different architectures in each region.

[0080] The method can utilize the recording files used to test the optimized offline speech recognition engine and the recognition results of the recording files to test the optimized offline speech recognition engines of different architectures in each region, and obtain the recognition accuracy of the optimized offline speech recognition engines of different architectures in each region, so as to subsequently calculate the load distribution ratio of the online speech recognition engine in each region according to the recognition accuracy of the optimized offline speech recognition engines of different architectures in each region.

[0081] In one embodiment of the present invention, the test samples are obtained from the recording platform of each region. For example, the test samples are obtained from the recording platform in Beijing, the test samples are obtained from the recording platform in Shanghai, the test samples are obtained from the recording platform in Hubei, the test samples are obtained from the recording platform in Heilongjiang, and so on, but not limited to this. In this embodiment, the recording files in the test samples can be audio files that are difficult to recognize, that is, when the recording files are recognized by the online speech recognition engine, the recognition results obtained are very different from the meaning expressed by the real user. The recognition results of the recording files in the above test samples are manually marked recognition results. Among them, the recording files can be recording files of various business scenarios.

[0082] refer to Figure 1 , through the training effect evaluation unit 140, using the recording file for testing the offline speech recognition engine and the recognition result of the recording file, the optimized offline speech recognition engines of different architectures in regions 1 to N are tested to obtain the recognition accuracy of the optimized offline speech recognition engines of different architectures in each region, so that the business load ratio of the engine with good optimization effect can be gradually increased through automated training and effect comparison. Specifically, the training effect evaluation unit 140 is used to adapt the interface of the offline speech recognition engine to realize the test of the offline recognition engine.

[0083] In one embodiment of the present invention, speech recognition engines of different architectures in various regions are tested in parallel using test samples, so that the comparison results can be obtained in real time and are more convincing to all manufacturers. In this embodiment, speech recognition engines of different architectures correspond to speech recognition engines of different manufacturers.

[0084] In one embodiment of the present invention, by Figure 3 The embodiment describes the optimization and Figure 4The offline speech recognition engine test described in the embodiment can detect manufacturers that are not fully optimized at any time through automated comparative testing in an environment with multiple speech recognition engines, so that manufacturers can be reminded in real time to train and optimize their own speech recognition engines.

[0085] Figure 5 A flowchart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0086] like Figure 5 As shown, the above method may further include steps S510 to S520.

[0087] In step S510, the load distribution ratio of the online speech recognition engine in each area is determined according to the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization of different architectures in each area.

[0088] In step S520, the load distribution ratio of the online speech recognition engine in each area is stored in a designated database.

[0089] The method can quickly and accurately determine the load distribution ratio of the online speech recognition engine in each area according to the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization of different architectures in each area, and store the load distribution ratio of the online speech recognition engine in each area in a designated database, so that it is convenient to update the load distribution ratio of the historical online speech recognition engine in the designated database according to the load distribution ratio of the current online speech recognition engine, thereby realizing dynamic adjustment of the load distribution ratio.

[0090] Figure 6 A flowchart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0091] like Figure 6 As shown, the above step S510 may specifically include step S610 and step S620.

[0092] In step S610, the specific proportion of the online speech recognition engine in allocating traffic is calculated according to the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization.

[0093] In step S620, the load distribution ratio of the online speech recognition engines in each area is determined according to the specific proportions of the multiple online speech recognition engines in each area in being responsible for allocating traffic.

[0094] The method can calculate the specific proportion of the online speech recognition engine when being responsible for allocating traffic according to the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization, and then determine the load distribution ratio of the online speech recognition engine in each area according to the specific proportion of multiple online speech recognition engines in each area when being responsible for allocating traffic. In this way, the load distribution ratio of the online speech recognition engine can be quickly and accurately calculated according to the recognition accuracy of the offline speech recognition engine after optimization in each area.

[0095] In one embodiment of the present invention, the specific proportion of the online speech recognition engine in allocating traffic is calculated based on the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization. For example, the specific proportion of the online speech recognition engine in allocating traffic is calculated by the following formula: m :

[0096] F m =A*(C n / C max )+B*(1-D n / C n )

[0097] Among them, F m is the specific proportion of the online speech recognition engine in allocating traffic, and m is the identifier of the online speech recognition engine. A and B are weight coefficients, which can be set according to the specific situation of the speech recognition engine manufacturer. n is the recognition accuracy of the optimized offline speech recognition engine with different architectures in each region, and n is the region identifier. max The maximum recognition accuracy in the optimized offline speech recognition engine for different architectures in each region. n The recognition accuracy of the offline speech recognition engine before optimizing different architectures in each region.

[0098] In one embodiment of the present invention, the load distribution ratio of the online speech recognition engines in each area is determined according to the specific proportion of the multiple online speech recognition engines in each area when they are responsible for allocating traffic. For example, area 1 contains three online speech recognition engines, namely online speech recognition engine A1, online speech recognition engine B1, and online speech recognition engine C1. The specific proportion of the online speech recognition engine A1 when it is responsible for allocating traffic F A1 , the specific proportion of online speech recognition engine B1 in allocating traffic F B1 , the specific proportion of online speech recognition engine C1 in allocating traffic F C1 Determine the load distribution ratio of the online speech recognition engine in each area as F A1 :FB1 : F C1 . For example, 25%:45%:30%.

[0099] Figure 7 A flowchart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0100] like Figure 7 As shown, the above method may further include step S710 and step S720.

[0101] In step S710, a first update task is received, wherein the first update task includes an identifier of an online speech recognition engine to be updated.

[0102] In step S720, the online speech recognition engine in each area is updated according to the identifier of the online speech recognition engine to be updated and the optimized offline speech recognition engine.

[0103] The method can update the online speech recognition engine in each area according to the identifier of the online speech recognition engine to be updated and the optimized offline speech recognition engine. In this way, the online speech recognition engine deployed in the online recognition environment can be updated in time, which can improve the recognition accuracy of the online speech recognition engine in the online recognition environment.

[0104] refer to Figure 1 After the optimization of the offline speech recognition engines of different architectures in each region is completed by the optimization unit 130, the optimized offline speech recognition engines of different architectures in each region can be obtained. In this case, a first update task can be generated based on the optimized offline speech recognition engine. The identifier of the online speech recognition engine to be updated in the first update task is the same as the identifier of the optimized offline speech recognition engine, that is, the speech recognition engines of the same manufacturer. For example, the identifier of the online speech recognition engine to be updated in the first update task is A1, and the identifier of the optimized offline speech recognition engine is A1. At this time, the online speech recognition engine A1 of the region in the online recognition environment can be updated based on the identifier of the online speech recognition engine to be updated being A1 and the optimized offline speech recognition engine being A1.

[0105] Figure 8 A flowchart of a method for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0106] like Figure 8 As shown, the above method may further include step S810 and step S820.

[0107] In step S810, a second update task is received, where the second update task includes a current load distribution ratio of the online speech recognition engine in each area.

[0108] In step S820, the historical load distribution ratio of the online speech recognition engine in each area of ​​the designated database is updated using the current load distribution ratio of the online speech recognition engine in each area.

[0109] The method can utilize the current load distribution ratio of the online speech recognition engine in each area to update the historical load distribution ratio of the online speech recognition engine in each area of ​​the above-mentioned specified database. In this way, the historical load distribution ratio of the online speech recognition engine can be updated in a timely manner, so that the load balancing of the online speech recognition engines with different architectures in various areas in the online recognition environment can be dynamically adjusted.

[0110] refer to Figure 1 After the optimized offline speech recognition engines of different architectures in each region are tested by the training effect evaluation unit 140, the recognition accuracy of the optimized offline speech recognition engines of different architectures in each region can be obtained. According to the recognition accuracy of the optimized offline speech recognition engines of different architectures in each region, the load distribution ratio of the online speech recognition engines of different architectures in each region can be determined, that is, the current load distribution ratio of the online speech recognition engines. In this case, a second update task can be generated according to the current load distribution ratio of the online speech recognition engines. In response to the second update task, the current load distribution ratio of the online speech recognition engines in each region can be used to update the historical load distribution ratio of the online speech recognition engines in each region of the specified database. For example, the second update task includes the current load distribution ratio of the online speech recognition engines in region 1. In response to the second update task, the current load distribution ratio of the online speech recognition engines in region 1 can be used to update the historical load distribution ratio of the online speech recognition engines in region 1 of the specified database.

[0111] Fig. 9 The block diagram of a device for load balancing of heterogeneous speech recognition engines according to an embodiment of the present invention is schematically shown.

[0112] like Fig. 9 As shown, the above-mentioned device 900 for load balancing of heterogeneous speech recognition engines may include a first receiving module 901, a first acquiring module 902 and an allocating module 903.

[0113] Specifically, the first receiving module 901 is used to receive speech recognition requests from different areas.

[0114] The first acquisition module 902 is used to obtain the load distribution ratio of the online speech recognition engine in each area according to the area identifier in the speech recognition request of the different areas, wherein online speech recognition engines with different architectures are deployed in each area, and the load distribution ratio of the online speech recognition engine in each area is determined based on the recognition accuracy of the offline speech recognition engines with different architectures in the area during the optimization process.

[0115] The allocation module 903 is used to allocate the speech recognition requests in different areas according to the load distribution ratio of the online speech recognition engine in each area.

[0116] The device 900 for load balancing heterogeneous speech recognition engines can receive speech recognition requests from different regions, and obtain the load distribution ratio of the online speech recognition engine in each region according to the region identifiers in the speech recognition requests from different regions, wherein online speech recognition engines with different architectures are deployed in each region, and the load distribution ratio of the online speech recognition engines in each region is determined based on the recognition accuracy of offline speech recognition engines with different architectures in the region during the optimization process, and then the speech recognition requests from different regions are distributed according to the load distribution ratio of the online speech recognition engines in each region. In this way, the speech recognition requests from different regions are distributed according to the speech recognition capabilities of each engine, thereby achieving load balancing of each speech recognition engine in a practical sense.

[0117] According to an embodiment of the present invention, the device 900 for load balancing of heterogeneous speech recognition engines can be used to implement Figure 2 The embodiment describes a method for load balancing of heterogeneous speech recognition engines.

[0118] Fig.10 A block diagram of an apparatus for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0119] like Fig.10 As shown, the above-mentioned device 900 for load balancing of heterogeneous speech recognition engines may further include a second acquisition module 904 and a training module 905 .

[0120] Specifically, the second acquisition module 904 is used to acquire training samples from the recording platform of each area, wherein the training samples include recording files for training the offline speech recognition engine and the recognition results of the recording files.

[0121] The training module 905 optimizes the offline speech recognition engine of different architectures in each region by using the recording file used for training the offline speech recognition engine and the recognition result of the recording file to obtain an optimized offline speech recognition engine.

[0122] The heterogeneous speech recognition engine load balancing device 900 can use the recording files used to train the offline speech recognition engine and the recognition results of the recording files to optimize the offline speech recognition engines of different architectures in each area to obtain the optimized offline speech recognition engine. This can improve the recognition ability of the offline speech recognition engine and make the recognition accuracy of the optimized offline speech recognition engine higher.

[0123] According to an embodiment of the present invention, the device 900 for load balancing of heterogeneous speech recognition engines can be used to implement Figure 3 The embodiment describes a method for load balancing of heterogeneous speech recognition engines.

[0124] Fig.11 A block diagram of an apparatus for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0125] like Fig.11 As shown, the above-mentioned device 900 for load balancing of heterogeneous speech recognition engines may further include a third acquisition module 906 and a testing module 907 .

[0126] Specifically, the third acquisition module 906 is used to obtain a test sample from the recording platform of each area, wherein the test sample includes a recording file for testing the optimized offline speech recognition engine and a recognition result of the recording file.

[0127] The testing module 907 uses the recording file used to test the optimized offline speech recognition engine and the recognition result of the recording file to test the optimized offline speech recognition engine with different architectures in each area to obtain the recognition accuracy of the optimized offline speech recognition engine with different architectures in each area.

[0128] The heterogeneous speech recognition engine load balancing device 900 can use the recording file used to test the optimized offline speech recognition engine and the recognition result of the recording file to test the optimized offline speech recognition engine with different architectures in each area, and obtain the recognition accuracy of the optimized offline speech recognition engine with different architectures in each area, so as to calculate the load distribution ratio of the online speech recognition engine in each area according to the recognition accuracy of the optimized offline speech recognition engine with different architectures in each area.

[0129] According to an embodiment of the present invention, the device 900 for load balancing of heterogeneous speech recognition engines can be used to implement Figure 4 The embodiment describes a method for load balancing of heterogeneous speech recognition engines.

[0130] Fig.12 A block diagram of an apparatus for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0131] like Fig.12 As shown, the above-mentioned device 900 for load balancing of heterogeneous speech recognition engines may also include a distribution ratio determination module 908 and a storage module 909.

[0132] Specifically, the distribution ratio determination module 908 is used to determine the load distribution ratio of the online speech recognition engine in each area according to the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization of different architectures in each area.

[0133] The storage module 909 is used to store the load distribution ratio of the online speech recognition engine in each area in a designated database.

[0134] The device 900 for load balancing heterogeneous speech recognition engines can quickly and accurately determine the load distribution ratio of the online speech recognition engine in each area based on the recognition accuracy of the offline speech recognition engines after optimization and the recognition accuracy of the offline speech recognition engines before optimization of different architectures in each area, and store the load distribution ratio of the online speech recognition engines in each area in a designated database, so as to facilitate the subsequent update of the load distribution ratio of the historical online speech recognition engines in the designated database according to the load distribution ratio of the current online speech recognition engine, thereby realizing dynamic adjustment of the load distribution ratio.

[0135] According to an embodiment of the present invention, the device 900 for load balancing of heterogeneous speech recognition engines can be used to implement Figure 5 The embodiment describes a method for load balancing of heterogeneous speech recognition engines.

[0136] Fig.13 A block diagram of an apparatus for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0137] like Fig.13 As shown, the above-mentioned allocation ratio determination module 908 can specifically include an allocation ratio determination module 908-1 and an allocation ratio determination sub-module 908-2.

[0138] Specifically, the allocation ratio determination module 908-1 is used to calculate the specific ratio of the online speech recognition engine in allocating traffic according to the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization.

[0139] The distribution ratio determination submodule 908-2 is used to determine the load distribution ratio of the online speech recognition engines in each area according to the specific proportions of the multiple online speech recognition engines in each area when being responsible for distributing traffic.

[0140] The allocation ratio determination module 908 can calculate the specific proportion of the online speech recognition engine in allocating traffic based on the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization, and then determine the load distribution ratio of the online speech recognition engine in each area based on the specific proportion of the multiple online speech recognition engines in each area in allocating traffic. In this way, the load distribution ratio of the online speech recognition engine can be quickly and accurately calculated based on the recognition accuracy of the offline speech recognition engine after optimization in each area.

[0141] According to an embodiment of the present invention, the allocation ratio determination module 908 can be used to implement Figure 6 The embodiment describes a method for load balancing of heterogeneous speech recognition engines.

[0142] Fig.14 A block diagram of an apparatus for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0143] like Fig.14 As shown, the above-mentioned device 900 for load balancing of heterogeneous speech recognition engines may also include a second receiving module 910 and a first updating module 911.

[0144] Specifically, the second receiving module 910 is used to receive a first update task, where the first update task includes an identifier of an online speech recognition engine to be updated.

[0145] The first updating module 911 is used to update the online speech recognition engine in each area according to the identifier of the online speech recognition engine to be updated and the optimized offline speech recognition engine.

[0146] The heterogeneous speech recognition engine load balancing device 900 can update the online speech recognition engine in each area according to the identifier of the online speech recognition engine to be updated and the optimized offline speech recognition engine. In this way, the online speech recognition engine deployed in the online recognition environment can be updated in time, which can improve the recognition accuracy of the online speech recognition engine in the online recognition environment.

[0147] According to an embodiment of the present invention, the device 900 for load balancing of heterogeneous speech recognition engines can be used to implement Figure 7 The embodiment describes a method for load balancing of heterogeneous speech recognition engines.

[0148] Fig.15 A block diagram of an apparatus for load balancing of heterogeneous speech recognition engines according to another embodiment of the present invention is schematically shown.

[0149] like Fig.15As shown, the above-mentioned device 900 for load balancing of heterogeneous speech recognition engines may further include a third receiving module 912 and a second updating module 913 .

[0150] Specifically, the third receiving module 912 is used to receive a second update task, where the second update task includes a current load distribution ratio of the online speech recognition engine in each area.

[0151] The second updating module 913 updates the historical load distribution ratio of the online speech recognition engine in each area of ​​the designated database by using the current load distribution ratio of the online speech recognition engine in each area.

[0152] The device 900 for load balancing of heterogeneous speech recognition engines can utilize the current load distribution ratio of the online speech recognition engines in each area to update the historical load distribution ratio of the online speech recognition engines in each area of ​​the above-mentioned specified database. In this way, the historical load distribution ratio of the online speech recognition engines can be updated in a timely manner, so that the load balancing of online speech recognition engines with different architectures in various areas in the online recognition environment can be dynamically adjusted.

[0153] According to an embodiment of the present invention, the device 900 for load balancing of heterogeneous speech recognition engines can be used to implement Figure 8 The embodiment describes a method for load balancing of heterogeneous speech recognition engines.

[0154] Since each module of the apparatus 900 for load balancing heterogeneous speech recognition engines according to the exemplary embodiment of the present invention can be used to implement the above 2~ Figure 8 The steps of an example embodiment of the method for load balancing of heterogeneous speech recognition engines are described. Therefore, for details not disclosed in the device embodiment of the present invention, please refer to the embodiment of the method for load balancing of heterogeneous speech recognition engines described above.

[0155] It can be understood that the first receiving module 901, the first acquisition module 902, the allocation module 903, the second acquisition module 904, the training module 905, the third acquisition module 906, the testing module 907, the allocation ratio determination module 908, the allocation ratio determination module 908-1, the allocation ratio determination submodule 908-2, the storage module 909, the second receiving module 910, the first update module 911, the third receiving module 912, and the second update module 913 can be combined in one module, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present invention, at least one of the first receiving module 901, the first acquisition module 902, the allocation module 903, the second acquisition module 904, the training module 905, the third acquisition module 906, the testing module 907, the allocation ratio determination module 908, the allocation ratio determination module 908-1, the allocation ratio determination sub-module 908-2, the storage module 909, the second receiving module 910, the first update module 911, the third receiving module 912, and the second update module 913 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware such as any other reasonable way of integrating or packaging circuits, or in an appropriate combination of software, hardware and firmware. Alternatively, at least one of the first receiving module 901, the first acquisition module 902, the allocation module 903, the second acquisition module 904, the training module 905, the third acquisition module 906, the testing module 907, the allocation ratio determination module 908, the allocation ratio determination module 908-1, the allocation ratio determination sub-module 908-2, the storage module 909, the second receiving module 910, the first update module 911, the third receiving module 912, and the second update module 913 can be at least partially implemented as a computer program module, and when the program is run by a computer, the functions of the corresponding module can be executed.

[0156] Reference below Fig.16 , which shows a schematic diagram of the structure of a computer system 1600 of an electronic device suitable for implementing an embodiment of the present invention. Fig.16 The computer system 1600 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0157] like Fig.16As shown, the computer system 1600 includes a central processing unit (CPU) 1601, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1602 or the program loaded from the storage part 1608 to the random access memory (RAM) 1603. In the RAM 1603, various programs and data required for system operation are also stored. The CPU 1601, the ROM 1602, and the RAM 1603 are connected to each other through a bus 1604. An input / output (I / O) interface 1605 is also connected to the bus 1604.

[0158] The following components are connected to the I / O interface 1605: an input section 1606 including a keyboard, a mouse, etc.; an output section 1607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1608 including a hard disk, etc.; and a communication section 1609 including a network interface card such as a LAN card, a modem, etc. The communication section 1609 performs communication processing via a network such as the Internet. A drive 1610 is also connected to the I / O interface 1605 as needed. A removable medium 1611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1610 as needed, so that a computer program read therefrom is installed into the storage section 1608 as needed.

[0159] 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 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 network through a communication section 1609, and / or installed from a removable medium 1611. When the computer program is executed by a central processing unit (CPU) 1601, the above-mentioned functions defined in the system of the present application are executed.

[0160] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may 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 may 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 may 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 may 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 may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, 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 appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0161] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0162] The units involved in the embodiments of the present invention may be implemented by software or hardware, and the units described may also be arranged in a processor. The names of these units do not, in some cases, limit the units themselves.

[0163] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method for load balancing of heterogeneous speech recognition engines as described in the above embodiment.

[0164] For example, the electronic device can implement Figure 2 As shown in: In step S210, speech recognition requests from different regions are received. In step S220, the load distribution ratio of the online speech recognition engine in each region is obtained according to the region identifier in the speech recognition requests from the different regions, wherein online speech recognition engines of different architectures are deployed in each region, and the load distribution ratio of the online speech recognition engine in each region is determined based on the recognition accuracy of offline speech recognition engines of different architectures in the region during the optimization process. In step S230, the speech recognition requests from the different regions are distributed according to the load distribution ratio of the online speech recognition engines in each region.

[0165] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0166] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation of the present invention.

[0167] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

[0168] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for load balancing of heterogeneous speech recognition engines, characterized in that: The method includes: Receive speech recognition requests from different regions; Obtaining, according to the region identifiers in the speech recognition requests of the different regions, a load distribution ratio of the online speech recognition engine in each region, wherein online speech recognition engines of different architectures are deployed in each region, and the load distribution ratio of the online speech recognition engine in each region is determined based on the recognition accuracy of the offline speech recognition engines of different architectures in the region during the optimization process; The speech recognition requests in the different regions are distributed according to the load distribution ratio of the online speech recognition engine in each region.

2. The method according to claim 1, characterized in that The method further includes: Acquire training samples from the recording platform of each region, wherein the training samples include recording files for training an offline speech recognition engine and recognition results of the recording files; The recording file used for training the offline speech recognition engine and the recognition result of the recording file are used to optimize the offline speech recognition engine of different architectures in each area to obtain an optimized offline speech recognition engine.

3. The method according to claim 2, characterized in that The method further includes: Acquire a test sample from the recording platform of each region, wherein the test sample includes a recording file for testing the optimized offline speech recognition engine and a recognition result of the recording file; The optimized offline speech recognition engine of different architectures in each region is tested by using the recording file used to test the optimized offline speech recognition engine and the recognition result of the recording file to obtain the recognition accuracy of the optimized offline speech recognition engine of different architectures in each region.

4. The method according to claim 3, characterized in that The method further includes: Determine the load distribution ratio of the online speech recognition engine in each region based on the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization of different architectures in each region; The load distribution ratio of the online speech recognition engine in each region is stored in a specified database.

5. The method according to claim 4, characterized in that According to the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization of different architectures in each region, the load distribution ratio of the online speech recognition engine in each region is determined, including: Calculate the specific proportion of the online speech recognition engine in allocating traffic based on the recognition accuracy of the offline speech recognition engine after optimization and the recognition accuracy of the offline speech recognition engine before optimization. The load distribution ratio of the online speech recognition engines in each region is determined based on the specific proportions of the multiple online speech recognition engines in each region in distributing traffic.

6. The method according to claim 2, characterized in that The method further includes: receiving a first update task, wherein the first update task includes an identifier of an online speech recognition engine to be updated; The online speech recognition engine in each area is updated according to the identifier of the online speech recognition engine to be updated and the optimized offline speech recognition engine.

7. The method according to claim 4, characterized in that The method further includes: receiving a second update task, wherein the second update task includes a current load distribution ratio of the online speech recognition engine in each region; The historical load distribution ratio of the online speech recognition engine in each area of ​​the designated database is updated by using the current load distribution ratio of the online speech recognition engine in each area.

8. A device for load balancing of heterogeneous speech recognition engines, characterized in that: The device includes: A first receiving module, used for receiving speech recognition requests from different regions; A first acquisition module is used to acquire the load distribution ratio of the online speech recognition engine in each area according to the area identifier in the speech recognition request of the different areas, wherein online speech recognition engines of different architectures are deployed in each area, and the load distribution ratio of the online speech recognition engine in each area is determined based on the recognition accuracy of the offline speech recognition engines of different architectures in the area during the optimization process; The allocation module is used to allocate the speech recognition requests in the different areas according to the load distribution ratio of the online speech recognition engine in each area.

9. An electronic device, comprising: one or more processors; as well as A storage device, used for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the method according to any one of claims 1 to 7.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.