Speech recognition method, device, electronic device, and readable storage medium

Abstract

The application discloses a speech recognition method and device, an electronic device and a readable storage medium. The method comprises the following steps: acquiring speech information and auxiliary information; wherein the auxiliary information comprises user information; performing speech recognition on the speech information to obtain at least two initial recognition results; and determining a final speech recognition result corresponding to the speech information from the at least two initial recognition results according to the auxiliary information. In this way, the application can assist in recognizing the initial recognition results by using the auxiliary information when there are multiple recognition results in the speech information recognition, so as to correctly recognize the multiple initial recognition results corresponding to the speech information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of speech recognition, in particular to a speech recognition method, device, electronic device and readable storage medium. BACKGROUND

[0002] Generally, as people's various demands for product equipment usage increase, when using product equipment for voice control, users often hope to maintain the timeliness of product equipment control and the accuracy of voice in product equipment.

[0003] Generally, in the field of speech recognition technology, in order to use product equipment accurately, a sound pickup device is often provided on the product equipment, and the user can control and interact with the application program or web interface by speaking to the sound pickup device through the application program or web interface of the mobile terminal and other product equipment.

[0004] At present, when a short sentence is received, the product equipment often has the problem of homonym recognition difficulty for short text recognition, such as the same pronunciation or no distinction between front and back nasal sounds, so that the product equipment cannot accurately obtain the real intention of the user, and further makes the Automatic Speech Recognition (ASR) error and cannot distinguish the user's intention. SUMMARY

[0005] The first aspect of the embodiment of the present application provides a speech recognition method, which comprises: obtaining voice information and auxiliary information; wherein the auxiliary information comprises user information; performing speech recognition on the voice information to obtain at least two initial recognition results; and determining a final speech recognition result corresponding to the voice information from the at least two initial recognition results according to the auxiliary information.

[0006] The second aspect of the embodiment of the present application provides a speech recognition device, which comprises: an obtaining module for obtaining voice information and auxiliary information; wherein the auxiliary information comprises user information; an identification module for performing speech recognition on the voice information to obtain at least two initial recognition results; and a determination module for determining a final speech recognition result corresponding to the voice information from the at least two initial recognition results according to the auxiliary information.

[0007] The third aspect of the embodiment of the present application provides an electronic device, which comprises: a processor and a memory, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the method of the first aspect.

[0008] The fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program can be executed by a processor to implement the method of the first aspect of the present application.

[0009] The beneficial effect of the present application is that the at least two initial recognition results converted by using the auxiliary information can be distinguished in the process of recognizing the voice information, so that the multiple initial recognition results corresponding to the voice information can be correctly recognized, and the homonym short text recognition can be distinguished and screened, thereby obtaining the final voice recognition result corresponding to the voice information. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0011] Figure 1 is a flowchart of an embodiment of the voice recognition method of the present application;

[0012] Figure 2 is a flowchart of an embodiment of the voice recognition method of the present application Figure 1

[0013] Figure 3 is a flowchart of an embodiment of the voice recognition method of the present application Figure 2

[0014] Figure 4 is a flowchart of an embodiment of the voice recognition method of the present application Figure 3

[0015] Figure 5 is a flowchart of an embodiment of the voice recognition method of the present application Figure 4

[0016] Figure 6 is a flowchart of another embodiment of the voice recognition method of the present application;

[0017] Figure 7 is a structural block diagram of a voice recognition device provided by an embodiment of the present application;

[0018] Figure 8 is a structural schematic diagram of an electronic device provided by an embodiment of the present application;

[0019] Figure 9 is a structural schematic diagram of a computer readable storage medium provided by the present application;

[0020] Figure 10 is a schematic block diagram of the hardware architecture of the terminal. DETAILED DESCRIPTION ​​​​

[0021] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0022] It is to be understood that the terminology "includes", "has", "holds", "contains" and / or "comprising", when used in this specification and in the following claims, indicates the presence of the described features, integers, steps, operations, elements, and / or components but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0023] It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used in this specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0024] It will be further understood that the terms "and", "or", as used herein, refer to a "and / or", unless otherwise indicated by the context. In other words, the terms "and" and "or" as used herein, unless otherwise indicated by the context, can be construed to both cover every possibility of a combination and also each possibility that exists alone.

[0025] As used in this specification and the appended claims, the term "if' can be construed to mean "when" or "once" or "in response to determining" or "in response to detecting", depending on the context. In other words, the phrase "if [a given condition or event occurs]" as used herein can be interpreted to mean "once [the given condition or event occurs]" or "in response to determining [the given condition or event]" or "in response to detecting [the given condition or event]", depending on the context.

[0026] In order to illustrate the technical solutions of the present application, the present application provides a speech recognition method, please refer to Figure 1 , Figure 1 is the overall system framework schematic diagram of the speech recognition method of the present application, which specifically comprises the following steps:

[0027] S11: obtaining speech information and auxiliary information;

[0028] Speech recognition technology is an extremely important capability in artificial intelligence, especially for electronic devices such as mobile phones, sound boxes, car machines, etc. equipped with intelligent voice assistants. Generally speaking, when a user issues a voice instruction, the electronic device can obtain the speech information corresponding to the voice instruction.

[0029] At present, speech correction model is often used to recognize and correct the obtained speech data, however, it cannot well correct short text content, for example, if the pinyin corresponding to the speech issued by the user is: mianyang, the electronic device cannot understand whether the user is asking the text content of "mianyang" or "mianyang", and cannot query the encyclopedia result.

[0030] For another example, if the user instruction is: open bailu application program, there are two audio products "bailu" and "bailu" on the market, at this time the electronic device also cannot identify the text content corresponding to the short speech represented by the homonym.

[0031] Therefore, the electronic device in the embodiment can also obtain auxiliary information for assisting in identifying the speech information, wherein the auxiliary information includes user information, so as to distinguish other conversion homonyms of the speech information.

[0032] S12: identifying the speech information to obtain at least two initial recognition results;

[0033] Generally, the smart voice assistant is installed on the electronic device, the speech information is identified through the smart voice assistant, and at least two initial recognition results can be identified through the database pre-stored in the electronic device in the conversion process of the homonyms of the speech information.

[0034] The algorithm of the smart voice assistant can understand the intention of natural language (NLU), and then execute the user's requirement according to the understanding of the user instruction, and then broadcast the prompt through voice synthesis.

[0035] As can be seen, speech recognition is the first step and the most important step for the voice assistant to execute the user instruction. Only when the speech recognition is accurate and can understand the user's intention to the greatest extent, the subsequent execution of the command for the user can be better.

[0036] S13: determining the final speech recognition result of the speech information from the at least two initial recognition results according to the auxiliary information.

[0037] Generally, when the user speaks to the electronic device, the speech to be identified usually represents only one intention of the user, and the final speech recognition result converted by the short speech usually has only one short word.

[0038] Because the auxiliary information includes user information, it can better associate the real intention of the user, so the final speech recognition result can be determined from the at least two initial recognition results according to the auxiliary information. Specifically, for example, the final speech recognition result can be determined from the at least two initial recognition results according to the auxiliary information.

[0039] For example, the real intention of the user is determined by auxiliary information such as user positioning, historical information, etc. Of course, a hot word strategy can also be used, which refers to filtering the initial voice recognition result according to the hot search in the current network to reduce the misexecution of user instructions. For example, the weight of one of the at least two initial recognition results is determined to determine the final voice recognition result, thereby greatly improving the user experience, which is not limited here.

[0040] Therefore, the present application can use auxiliary information to identify at least two initial recognition results obtained by converting voice information to correctly identify multiple initial recognition results corresponding to voice information, further distinguish and screen homonymic short text recognition, and obtain the final voice recognition result corresponding to voice information.

[0041] Further, the final voice recognition result of the voice information is determined from the at least two initial recognition results according to the auxiliary information. Please refer to Figure 2 , Figure 2 is a flowchart of an embodiment of step S13 in the present application Figure 1 , specifically including the following steps:

[0042] S21: Determine the correlation degree of the initial recognition result and the auxiliary information;

[0043] Since the auxiliary information can represent the real intention of the user, before the real intention of the user is identified, it is necessary to determine the correlation degree of the voice information and the auxiliary information, that is, to determine the correlation degree of the initial recognition result and the auxiliary information.

[0044] Among them, the auxiliary information can include at least one of address information, application program information, user gender information, user age, and user application program preference information in a preset number of days. And the auxiliary information can be appropriately extended according to the time period when the user uses the mobile phone, which is not limited here.

[0045] S22: Determine the final voice recognition result from the at least two initial recognition results according to the correlation degree corresponding to each initial recognition result.

[0046] According to the conversion of the voice assistant, the at least two initial recognition results obtained are identified and confirmed to obtain the real intention of the user. Specifically, the final voice recognition result can be determined from the at least two initial recognition results according to the correlation degree corresponding to each initial recognition result.

[0047] Specifically, the degree of association between the initial recognition result and the auxiliary information also represents the real intention of the user. By the size of the degree of association corresponding to each initial recognition result, the final speech recognition result can be determined from at least two initial recognition results. For example, the greater the degree of association, the closer it is to the real intention of the user, and the smaller the degree of association, the farther it is from the real intention of the user. Therefore, the initial recognition result with a large degree of association can be determined as the final speech recognition result.

[0048] Further, the degree of association between the initial recognition result and the auxiliary information is determined. Please refer to Figure 3 , Figure 3 is a flowchart of an embodiment of step S21 in the present application Figure 2 , and specifically includes the following steps:

[0049] S31: performing feature extraction on the initial recognition result to obtain a first feature vector;

[0050] When the user issues a voice instruction, the speech assistant of the electronic device can use the text correction model before the final recognition content is obtained. By performing speech recognition on the voice information, i.e., performing speech recognition on the audio data, the corresponding speech recognition text can be obtained.

[0051] Thereafter, the initial recognition result converted from the voice data to be recognized is encoded, thereby obtaining a first feature vector corresponding to the initial recognition result, wherein the first feature vector can be marked by a plurality of classification labels.

[0052] The initial recognition result is subjected to feature extraction to obtain a first feature vector for representing the initial recognition result. The initial recognition result is not directly obtained because the data amount of the first feature vector is smaller than that of the initial recognition result, so that the storage space can be saved by extracting the first feature vector.

[0053] S32: performing feature extraction on the auxiliary information to obtain a second feature vector;

[0054] In order to better assist in confirming the initial recognition result, the second feature vector corresponding to the auxiliary information, such as the feature vector of the reference text of the user's personal information, can be cached in advance, so that the second feature vector can be obtained by performing feature extraction on the auxiliary information.

[0055] Specifically, the initial recognition result is a speech recognition text. After the user inputs the instruction, the second feature vector of the reference text of the user's personal information and the first feature vector of the speech recognition text can be input to the decoder for decoding.

[0056] S33: determining the similarity between the first feature vector and the second feature vector as the degree of association between the initial recognition result and the auxiliary information.

[0057] Specifically, by determining the similarity of the first feature vector and the second feature vector, the degree of association between the initial recognition result and the user personal information can be determined. Specifically, for example, the first feature vector represents a location, and the second feature vector represents a map. It is determined that the location and the map have a certain degree of similarity, and the degree of association between the location and the map is established.

[0058] In addition, after step S32, the operation is continued. According to the decoding result of the first feature vector and the decoding result of the second feature vector, the first confidence of two or more results can be calculated. When the first confidence is greater than a human-specified preset value, the similarity between the first feature vector and the second feature vector is within the preset distance range. Therefore, the degree of similarity between the initial recognition result and the auxiliary information can be determined by the size of the first confidence. Specifically, the degree of similarity between the initial recognition result and the auxiliary information is positively correlated with the first confidence.

[0059] Further, the similarity of the first feature vector and the second feature vector is determined. Please refer to Figure 4 , Figure 4 is a flowchart of a specific embodiment of step S33 in the present application Figure 3 , and specifically includes the following steps:

[0060] S41: Call the first feature vector to traverse the database to compare the second feature vector, wherein the second feature vector is pre-stored in the database;

[0061] In order to determine the similarity of the first feature vector and the second feature vector, first, the auxiliary information and the second feature vector corresponding to the auxiliary information can be stored in the database of the electronic device, such as each type of auxiliary information of location, animal, plant, etc. and the second feature vector corresponding to each type of auxiliary information.

[0062] Then, by calling the first feature vector to traverse the database to compare the second feature vector, the second feature vector is more likely to accurately match the second feature vector.

[0063] S42: Determine whether the first feature vector and the second feature vector are similar or the same class;

[0064] The same class means the same classification, for example, the first feature vector and the second feature vector are both strawberries, which are the same class. The similar class means the similar classification, for example, the first feature vector is strawberry, and the second feature vector is fruit, which is the similar class.

[0065] Specifically, by determining whether the first feature vector and the second feature vector are similar or the same class, it can be determined whether the first feature vector hits the second feature vector.

[0066] If the first feature vector and the second feature vector are similar or identical, it means that the first feature vector and the second feature vector have at least one identical label. Then proceed to step S43, which means that the first feature vector matches the second feature vector, and calculate the similarity between the first feature vector and the second feature vector.

[0067] Furthermore, to calculate the similarity between the first and second feature vectors, please refer to [link to relevant documentation]. Figure 5 , Figure 5 This application Figure 4 Step S43 is a flowchart of a specific embodiment, which specifically includes the following steps:

[0068] S51: Extract semantic labels from the first feature vector;

[0069] Depend on Figure 3 Step S31 reveals that the first feature vector can have multiple classification labels. By extracting the classification labels from the first feature vector, it is equivalent to obtaining the classification represented by the speech recognition text, thereby performing multiple classifications on the first feature vector.

[0070] Specifically, the classification label may include a speech label. By extracting the semantic label from the first feature vector, the semantics represented by the speech recognition text can be obtained, so that the second feature vector corresponding to the semantics can be compared and matched in the future.

[0071] S52: Based on semantic labels, obtain the weight coefficients corresponding to the auxiliary information represented by the second feature vector;

[0072] Specifically, semantic labels are used to determine the semantics represented by the initial recognition result. The weight coefficients (a / b / c) corresponding to the auxiliary information are determined based on the semantics corresponding to the initial recognition result. For example, a can represent the weight coefficient of the address information corresponding to the user's preset number of days, b can represent the weight coefficient of the user's application preference information, and c can represent the weight coefficient of the user's age information.

[0073] These weighting coefficients are determined by the influence of the corresponding user information on the recognition, and can usually be set by human preset, so no limitation is made here.

[0074] S53: Based on auxiliary information and weighting coefficients, calculate the confidence level between the first feature vector and the second feature vector to determine the similarity between the first feature vector and the second feature vector.

[0075] Specifically, the confidence level is calculated as follows:

[0076] By defining the weight of each type of user information in step S52, it is assumed that the address information corresponding to the user's preset number of days (such as the places the user has been to in the past seven days) is u1, the user's application program preference information is u2, and the user's age information is u3. By analogy, the initial value of each type of user information is 1, indicating that the voice device has not been used yet. This is an artificial preset, and other values can also be preset, which is not limited here. In the subsequent use process, the user information here can be infinitely increased.

[0077] The possible value S of a certain homonym recognition result is a*u1+b*u2+c*u3+…, and a / b / c is the weight coefficient of the corresponding user information. The weight coefficient of this block is determined according to the influence of the corresponding user information on the recognition, and can be accumulated. For example, if the semantic hits these u, then these u exist, otherwise u is 0.

[0078] If there is no corresponding user information to assist in a certain recognition, the weight coefficient is 0. Therefore, as long as there is one or more homonym recognition results, each result can obtain a confidence level. The higher the confidence level, the closer the result is to the user's true intention.

[0079] For example, if the user issues the command "mianyang", it is difficult to determine whether it is recognized as "mianyang" or "mianyang". At this time, the user's location can be obtained to assist text recognition. If the user's location is in Sichuan Province, we will consider that the possibility of asking for a place (confidence) is higher than the confidence of asking for an animal. At this time, the electronic device will recognize it as Mianyang, and display the encyclopedia result of Mianyang to the user.

[0080] For example, if it is determined that the semantic of "mianyang" is address information, the weights a1=0.5, b1=0.9, and c1=0.8 are obtained for the address information, the application program information u2=0.6, and the user age information u3=0.6. Then S1=0.5*0.5(Mianyang)+0.9*0.6(map)+0.8*0.6(25 years old)=1.27; If it is determined that the semantic of "mianyang" is animal information, the weights a2=0.6, b2=0.4, and c2=0.7 are obtained, u1=0.5, u2=0.6, and u3=0.6 (for the same device), then S2=0.6*0.5(Mianyang)+0.4*0.6(map)+0.7*0.6(25 years old)=0.96; Because S1>S2, it is determined that the recognized is the address of Mianyang.

[0081] Further, please refer to Figure 6 , Figure 6 is a flowchart of another embodiment of the voice recognition method of the present application, which specifically comprises the following steps:

[0082] S61: determining whether the confidence degrees corresponding to the at least two initial recognition results are same;

[0083] If the confidence degrees are same, entering step S62, that is, obtaining the hot words corresponding to each initial recognition result, and comparing the hot degrees of the hot words to determine the final speech recognition result, so that the superiority of artificial intelligence is highlighted, and the user instruction misexecution is reduced. If the confidence degrees are different, entering step S63, that is, selecting the final speech recognition result according to the confidence degrees.

[0084] In addition, the user's mobile phone APP installation information can also be used to assist in judgment. If the user installs one of them, it is considered that the user has installed the app with higher confidence, and the speech recognition can accurately display the user's needs.

[0085] Therefore, the present application fills the blank of the current intelligent strategy of short text recognition, uses more accurate and high correlation user personal information to assist text content recognition, not only correctly displays the text recognition to the user in the user's UI interface, but also helps the correct execution of subsequent user instructions, which can greatly improve the user experience and increase the user's satisfaction with the product.

[0086] In addition, the present application also provides a speech recognition device, please refer to Figure 7 , Figure 7 is a structural block diagram of the speech recognition device provided by the present application. The speech recognition device 60 comprises an acquisition module 61, a recognition module 62 and a determination module 63.

[0087] The acquisition module 61 is used to acquire speech information and auxiliary information; wherein the auxiliary information comprises user information, at least one of address information of a preset number of days, application program information, user gender information, user age and user application program preference information;

[0088] The recognition module 62 is connected with the acquisition module 61, and is used to recognize the speech information to obtain at least two initial recognition results;

[0089] The determination module 63 is used to determine the final speech recognition result from the at least two initial recognition results according to the auxiliary information.

[0090] Therefore, in the process of recognizing the speech information, the present application can use the auxiliary information to distinguish the at least two initial recognition results obtained by conversion, and use the acquisition module 61 to acquire the user information included in the auxiliary information, so that the recognition module 62 correctly recognizes the multiple initial recognition results corresponding to the speech information, further distinguishes and filters the homonym short text recognition, and finally obtains the final speech recognition result corresponding to the speech information.

[0091] In an embodiment, the determining module 63 is connected to the identifying module 62, and is further configured to determine a relevance between the initial recognition result and the auxiliary information; and determine the final speech recognition result from the at least two initial recognition results according to the relevance corresponding to each initial recognition result.

[0092] In an embodiment, the system further comprises a processing module (not shown in the figure), configured to perform feature extraction on the initial recognition result to obtain a first feature vector, and perform feature extraction on the auxiliary information to obtain a second feature vector; and the determining module 63 is further configured to determine a similarity between the first feature vector and the second feature vector as the relevance between the initial recognition result and the auxiliary information. The processing module comprises the obtaining module 61, the identifying module 62 and the determining module 63.

[0093] In an embodiment, the processing module is further configured to call the first feature vector to traverse a database to compare the second feature vector, wherein the second feature vector is pre-stored in the database; determine whether the first feature vector and the second feature vector are of a similar class or the same class; if the first feature vector and the second feature vector are of a similar class or the same class, it means that the first feature vector hits the second feature vector, and a similarity between the first feature vector and the second feature vector is calculated.

[0094] In an embodiment, the processing module is further configured to extract a semantic label in the first feature vector; obtain a weight coefficient corresponding to the auxiliary information represented by the second feature vector based on the semantic label; and calculate a first confidence between the first feature vector and the second feature vector based on the auxiliary information and the weight coefficient, to determine the similarity between the first feature vector and the second feature vector. Determine whether second confidences corresponding to the at least two initial recognition results are the same; if the second confidences are the same, obtain a hot word corresponding to each initial recognition result, and compare hotness values of the hot words, and the hot word with a larger hotness value is determined as the final speech recognition result.

[0095] In addition, the present application also provides an electronic device, please refer to Figure 8 , Figure 8 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. The electronic device 70 comprises a processor 71 and a memory 72, the memory 72 stores a computer program 721, and the processor 71 is configured to execute the computer program 721 to perform the following flow:

[0096] obtain speech information and auxiliary information; wherein the auxiliary information comprises user information; perform recognition on the speech information to obtain at least two initial recognition results; and determine a final speech recognition result corresponding to the speech information from the at least two initial recognition results according to the auxiliary information.

[0097] In an implementation, the processor 71, when determining the final speech recognition result from the at least two initial recognition results according to the auxiliary information, can determine a relevance of the initial recognition result to the auxiliary information, and determine the final speech recognition result from the at least two initial recognition results according to the relevance of each initial recognition result.

[0098] In an implementation, the processor 71, when determining the relevance of the initial recognition result to the auxiliary information, can perform feature extraction on the initial recognition result to obtain a first feature vector, perform feature extraction on the auxiliary information to obtain a second feature vector, and determine a similarity of the first feature vector and the second feature vector as the relevance of the initial recognition result to the auxiliary information.

[0099] In an implementation, the processor 71, when determining the similarity of the first feature vector and the second feature vector, can perform traversal retrieval of the first feature vector on a database to compare the second feature vector, where the second feature vector is pre-stored in the database, determine whether the first feature vector and the second feature vector are of a similar class or the same class, and if so, it means that the first feature vector hits the second feature vector, and the similarity between the first feature vector and the second feature vector is calculated.

[0100] In an implementation, the processor 71, when calculating the similarity between the first feature vector and the second feature vector, can extract a semantic label in the first feature vector, obtain a weight coefficient corresponding to the auxiliary information represented by the second feature vector based on the semantic label, and calculate a confidence between the first feature vector and the second feature vector based on the auxiliary information and the weight coefficient to determine the similarity between the first feature vector and the second feature vector.

[0101] In an implementation, the processor 71 can further determine whether the second confidence corresponding to the at least two initial recognition results is the same, obtain a hot word corresponding to each initial recognition result if the second confidence is the same, and compare the hotness values of the hot words to determine the final speech recognition result.

[0102] In addition, the present application also provides a computer readable storage medium, please refer to Figure 9 , Figure 9 is a structural diagram of the computer readable storage medium provided by the present application, the computer readable storage medium 80 stores a computer program 81, the computer program 81 can be executed by the processor to realize the method as described above, which will not be repeated here.

[0103] Please refer to Figure 10 , Figure 10is a schematic block diagram of a hardware architecture of a terminal of the present application. The electronic device 900 can be a smart television, an industrial computer, a tablet computer, a mobile phone, a notebook computer, etc. The present embodiment takes the mobile phone as an example. The structure of the terminal 900 can include radio frequency (RF) circuit 910, memory 920, input unit 930, display unit 940, sensor 950, audio circuit 960, WiFi (wireless fidelity) module 970, processor 980, and power supply 990, etc. The RF circuit 910, memory 920, input unit 930, display unit 940, sensor 950, audio circuit 960, and WiFi module 970 are respectively connected with the processor 980; and the power supply 990 is used to provide power for the entire electronic device 900.

[0104] Specifically, the RF circuit 910 is used to send and receive signals; the memory 920 is used to store data, instruction information, etc.; the input unit 930 is used to input information, and can specifically include a touch panel 931 and other input devices 932 such as operation keys, etc.; the display unit 940 can include a display panel, etc.; the sensor 950 includes an infrared sensor, a laser sensor, etc., and is used to detect user proximity signals, distance signals, etc.; the speaker 961 and microphone 962 are connected with the processor 980 through the audio circuit 960, and are used to send and receive sound signals; the WiFi module 970 is used to receive and transmit WiFi signals; and the processor 980 is used to process data information of the mobile phone.

[0105] The above only describes some embodiments of the present application, and does not limit the protection scope of the present application. Any equivalent device or equivalent flow conversion, or direct or indirect application in other related technical fields, which is made by using the content of the present application specification and drawings, is also included in the patent protection scope of the present application.

Claims

1. A voice recognition method, characterized by, The method comprises: acquiring voice information and auxiliary information; wherein the auxiliary information at least comprises address information of a preset number of days, application program information, user gender information, user age, and user preference information for using an application program; performing voice recognition on the voice information to obtain at least two initial recognition results; performing feature extraction on the initial recognition results to obtain a first feature vector; performing feature extraction on the auxiliary information to obtain a second feature vector; determining a similarity between the first feature vector and the second feature vector as an association degree between the initial recognition results and the auxiliary information; determining a final voice recognition result corresponding to the voice information from the at least two initial recognition results according to the association degree corresponding to each initial recognition result; wherein the determination of the similarity between the first feature vector and the second feature vector comprises: calling the first feature vector to traverse a database for comparison with the second feature vector, wherein the second feature vector is pre-stored in the database; determining whether the first feature vector and the second feature vector are similar or identical; if they are similar or identical, the following steps are performed: extracting a semantic label in the first feature vector; acquiring a weight coefficient corresponding to the auxiliary information represented by the second feature vector based on the semantic label, the weight coefficient being determined according to an influence of the corresponding auxiliary information on recognition; calculating a confidence degree between the first feature vector and the second feature vector based on the auxiliary information and the weight coefficient to determine the similarity between the first feature vector and the second feature vector.

2. The method of claim 1, wherein, The method further comprises: determining whether the confidence degrees corresponding to the at least two initial recognition results are the same; if they are the same, acquiring a hot word corresponding to each initial recognition result and comparing hotness values of the hot words to determine the final voice recognition result.

3. A voice recognition device, characterized by: an acquisition module configured to acquire voice information and auxiliary information; wherein the auxiliary information at least comprises address information of a preset number of days, application program information, user gender information, user age, and user preference information for using an application program; an identification module configured to perform voice recognition on the voice information to obtain at least two initial recognition results; a determination module configured to perform feature extraction on the initial recognition results to obtain a first feature vector; perform feature extraction on the auxiliary information to obtain a second feature vector; determine a similarity between the first feature vector and the second feature vector as an association degree between the initial recognition results and the auxiliary information; and determine a final voice recognition result corresponding to the voice information from the at least two initial recognition results according to the association degree corresponding to each initial recognition result; wherein the determination of the similarity between the first feature vector and the second feature vector comprises: calling the first feature vector to traverse a database for comparison with the second feature vector, wherein the second feature vector is pre-stored in the database; determining whether the first feature vector and the second feature vector are of a similar class or the same class; if they are of a similar class or the same class, then performing the following steps: extracting a semantic label from the first feature vector; based on the semantic label, obtaining a weight coefficient corresponding to the auxiliary information represented by the second feature vector, the weight coefficient being determined according to the influence of the corresponding auxiliary information on the identification; based on the auxiliary information and the weight coefficient, calculating a confidence degree between the first feature vector and the second feature vector to determine the similarity between the first feature vector and the second feature vector.

4. An electronic device, comprising: comprising: a processor and a memory, the memory storing a computer program, the processor being configured to execute the computer program to implement the method of claim 1 or 2.

5. A computer readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program can be executed by the processor to implement the method of claim 1 or 2.

Images