Speech recognition result correction method and device, electronic equipment, medium and vehicle

Abstract

The invention provides a voice recognition result correction method and device, electronic equipment, a medium and a vehicle, and relates to the technical field of vehicles. The method comprises the following steps: performing voice recognition on input audio information to obtain a first recognition result; in response to input abnormal feedback information, obtaining actual text information corresponding to the audio information; through the actual text information, the first recognition result is corrected to obtain the second recognition result, the second recognition result is the actual recognition result corresponding to the audio information, error results of voice recognition are corrected, the problem that homophones are difficult to distinguish in the voice recognition process is solved, and the accuracy of voice recognition is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of vehicle technology, and in particular to a method, device, electronic device, medium and vehicle for correcting speech recognition results. Background Art

[0002] With the rapid development of artificial intelligence technology, voice recognition is being used more and more widely. Voice recognition supports many common services in life. For example, in the automotive field, driving navigation can be performed through voice input.

[0003] However, the voice recognition methods in the related art have a low accuracy rate in recognizing voice commands, and lack sufficient ability to distinguish homophones that exist in large numbers in voice commands. For example, when saying the destination location through voice commands during navigation, problems may easily arise due to homophones, which may lead to position deviation if one is not careful. Summary of the invention

[0004] The present disclosure provides a speech recognition result correction method, device, electronic device, medium and vehicle.

[0005] According to a first aspect of the present disclosure, a method for correcting speech recognition results is provided, the method comprising: performing speech recognition on input audio information to obtain a first recognition result; obtaining actual text information corresponding to the audio information in response to input abnormal feedback information; and correcting the first recognition result through the actual text information to obtain a second recognition result, the second recognition result being the actual recognition result corresponding to the audio information.

[0006] In some embodiments, performing speech recognition on input audio information to obtain a first recognition result includes: recognizing the audio information through a first language model, and obtaining candidate recognition results corresponding to the audio information, wherein the candidate recognition results include the first recognition result.

[0007] In some embodiments, the first recognition result is corrected through actual text information to obtain a second recognition result, including: determining an abnormal word sequence in the first recognition result based on abnormal feedback information; determining an actual word sequence in the actual text information corresponding to the abnormal word sequence; based on the actual word sequence, correcting the first recognition result, the correction processing including at least one of a first correction processing, a second correction processing, a third correction processing, and a fourth correction processing.

[0008] In some embodiments, performing a first correction process on the first recognition result based on the actual word sequence includes: replacing the abnormal word sequence with the actual word sequence to correct the first recognition result.

[0009] In some embodiments, performing a second correction process on the first recognition result based on the actual word sequence includes: using a first weighting coefficient to increase the probability value of the actual word sequence in the candidate recognition result to correct the first recognition result.

[0010] In some embodiments, based on the actual word sequence, performing a third correction processing on the first recognition result includes: using personalized information to build a second language model, the personalized information includes the actual word sequence; recognizing audio information through the second language model to obtain the probability of the third recognition result; determining the second weighting coefficient and the third weighting coefficient; using the second weighting coefficient and the third weighting coefficient, weightedly fusing the probability of the third recognition result with the probability of the first recognition result to correct the first recognition result, the probability of the first recognition result being obtained through the first language model.

[0011] In some embodiments, the first language model is a statistical model, and based on the actual word sequence, the fourth correction processing of the first recognition result includes: adding the actual word sequence to the candidate recognition result of the first language model, or deleting the abnormal word sequence from the candidate recognition result of the first language model to correct the first recognition result.

[0012] In some embodiments, the method further includes: in response to abnormal feedback information corresponding to the second recognition result, performing correction processing on the second recognition result.

[0013] According to an embodiment of the present disclosure, a first recognition result is obtained by performing speech recognition on input audio information; in response to input abnormal feedback information, actual text information corresponding to the audio information is obtained; the first recognition result is corrected through the actual text information to obtain a second recognition result, which is the actual recognition result corresponding to the audio information, thereby correcting erroneous results of speech recognition, solving the problem of homophones being difficult to distinguish during speech recognition, and improving the accuracy of speech recognition.

[0014] According to a second aspect of the present disclosure, a speech recognition result correction device is provided, which includes: a recognition unit, used to perform speech recognition on input audio information to obtain a first recognition result; an acquisition unit, used to obtain actual text information corresponding to the audio information in response to input abnormal feedback information; and a processing unit, used to correct the first recognition result through the actual text information to obtain a second recognition result, which is the actual recognition result corresponding to the audio information.

[0015] According to an embodiment of the present disclosure, a first recognition result is obtained by performing speech recognition on input audio information; in response to input abnormal feedback information, actual text information corresponding to the audio information is obtained; the first recognition result is corrected through the actual text information to obtain a second recognition result, which is the actual recognition result corresponding to the audio information, thereby correcting erroneous results of speech recognition, solving the problem of homophones being difficult to distinguish during speech recognition, and improving the accuracy of speech recognition.

[0016] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0017] at least one processor; and

[0018] a memory communicatively connected to at least one processor; wherein,

[0019] The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable the at least one processor to perform the method of the first aspect described above.

[0020] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the method of the aforementioned first aspect.

[0021] According to a fifth aspect of the present disclosure, a vehicle is provided, comprising the speech recognition result correction device as described in the second aspect or the electronic device as described in the third aspect.

[0022] According to an embodiment of the present disclosure, a first recognition result is obtained by performing speech recognition on input audio information; in response to input abnormal feedback information, actual text information corresponding to the audio information is obtained; the first recognition result is corrected through the actual text information to obtain a second recognition result, which is the actual recognition result corresponding to the audio information, thereby correcting the erroneous results of speech recognition, solving the problem of homophones being difficult to distinguish during speech recognition, and improving the accuracy of speech recognition.

[0023] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0025] Figure 1 A flowchart of a method for correcting speech recognition results provided by an embodiment of the present disclosure;

[0026] Figure 2 A flowchart of a method for correcting speech recognition results provided by an embodiment of the present disclosure;

[0027] Figure 3 A schematic diagram of the structure of a speech recognition result correction device provided by an embodiment of the present disclosure;

[0028] Figure 4 A schematic block diagram of an exemplary electronic device 400 provided for an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0030] A speech recognition result correction method, device, electronic device, medium, and vehicle according to an embodiment of the present disclosure will be described below with reference to the accompanying drawings.

[0031] With the rapid development of artificial intelligence technology, voice recognition is being used more and more widely. Voice recognition supports many common services in life. For example, in the automotive field, driving navigation can be performed through voice input.

[0032] However, the speech recognition model in the speech recognition method in the related art relies on general information, and the modeling of the user's usage habits, vehicle conditions, the field to which the command belongs, etc. is poor, resulting in low accuracy of speech recognition, especially the lack of sufficient ability to distinguish the large number of homophones in the voice commands. For example, when saying the name of a community through voice commands during navigation, problems may easily arise due to homophones, which may lead to position deviation if not paid attention to.

[0033] In response to the problems of the above-mentioned related technologies, the voice recognition result correction method disclosed in the present invention is used to solve the problem of users in in-vehicle scenarios being unable to input and select accurately while driving. During the voice recognition process, the user's intention judgment is combined, and the feedback information made by the user's instructions and personal preferences are utilized to accurately recognize the content in the voice commands issued by the user, thereby improving the accuracy of voice recognition and the performance of the voice recognition model.

[0034] The speech recognition result correction method proposed in the present disclosure can be applied to devices with speech recognition functions, such as mobile phone terminals, smart home appliances, vehicles, etc. The application scenarios are not limited in the embodiments of the present disclosure. The present disclosure mainly takes the in-vehicle speech recognition scenario as an example.

[0035] The following describes in detail a method, device, electronic device, medium and vehicle for correcting speech recognition results proposed in the present disclosure in conjunction with the accompanying drawings.

[0036] Figure 1 The following is a flow chart of a method for correcting speech recognition results provided by an embodiment of the present disclosure. Figure 1 As shown, the method is applied to a vehicle, and the method comprises the following steps:

[0037] Step 101: Perform speech recognition on input audio information to obtain a first recognition result.

[0038] In one implementation of the present disclosure, the input audio information can be collected by a sound collection device of the vehicle, and the object of collection can be the driver or all persons currently within the collection range of the vehicle. The sound collection device can be a microphone or a handset.

[0039] Before performing speech recognition on the input audio information, the audio information is also pre-processed, that is, filtering, noise reduction, silence removal and other operations are performed. By pre-processing the audio information, some interfering factors such as noise, echo and different speaking speeds can be removed.

[0040] Speech recognition is a technology that converts human speech into machine-readable text. The first recognition result refers to the recognition result with the highest probability value among the candidate recognition results obtained by the first speech recognition model recognizing the audio information. Among them, the candidate recognition result refers to the recognition result whose probability value obtained by the speech recognition model recognizing the audio information is greater than a preset threshold.

[0041] Step 102: in response to the input abnormal feedback information, obtain the actual text information corresponding to the audio information.

[0042] In one implementation of the present disclosure, abnormal feedback information is feedback information obtained by identifying abnormal operations of the user. For example, when the vehicle recognizes the user's command and evaluates "incorrect recognition"; performs a secondary selection operation on the recognition results such as the navigation address and media name; interacts with the same content multiple times, etc., it can be determined that the current first recognition result is abnormal, that is, abnormal feedback information is obtained.

[0043] The actual text information refers to the correct text information corresponding to the current audio information extracted from the user's abnormal feedback information. For example, when the user evaluates the first recognition result as "incorrect recognition", the user directly inputs the actual text information or the user selects the actual text information multiple times in the operation interface.

[0044] Step 103: Correct the first recognition result using the actual text information to obtain a second recognition result, where the second recognition result is the actual recognition result corresponding to the audio information.

[0045] In one embodiment of the present disclosure, an actual word sequence in the actual text information is determined, and the actual word sequence is used to correct the abnormal word sequence in the first recognition result. Based on the actual word sequence, the first recognition result is corrected, and the correction process includes at least one of the first correction process, the second correction process, the third correction process, and the fourth correction process. The second recognition result refers to the actual recognition result corresponding to the audio information obtained after the correction process.

[0046] It is understandable that in the process of performing the correction process, one correction process can be used, or a combination of multiple correction processes can be used. The method is selected based on whether the current second recognition result meets the user's needs, and is not limited in this disclosure. In other words, if the second recognition result cannot meet the actual needs of the user, that is, there are still abnormal word sequences in the second recognition result, other correction processes can be used to correct it again until it meets the user's needs.

[0047] Therefore, according to an embodiment of the present disclosure, a first recognition result is obtained by performing speech recognition on input audio information; in response to input abnormal feedback information, the actual text information corresponding to the audio information is determined; the first recognition result is corrected through the actual text information to obtain a second recognition result, and the second recognition result is the actual recognition result corresponding to the audio information, thereby correcting the erroneous results of speech recognition, solving the problem of homophones being difficult to distinguish during speech recognition, and improving the accuracy of speech recognition.

[0048] Figure 2 A flowchart of a method for correcting speech recognition results provided in an embodiment of the present disclosure. Figure 2 based on Figure 1 In the embodiment shown, step 103 is further defined. Figure 2 In the embodiment shown, step 103 includes step 203, step 204, and step 205, and step 103 includes step 206. Figure 2 As shown, the method includes the following steps.

[0049] Step 201: Perform speech recognition on input audio information to obtain a first recognition result.

[0050] In an embodiment of the present disclosure, audio information is recognized by using a first language model to obtain candidate recognition results corresponding to the audio information, where the candidate recognition results include a first recognition result.

[0051] It can be understood that there can be multiple candidate recognition results, and the probability value corresponding to each candidate recognition result is the result of adding the probability values ​​of multiple word sequences corresponding to the current audio information. The probability values ​​of different recognition results are different, and the first recognition result is the recognition result with the highest probability value among the current candidate recognition results.

[0052] In the present disclosure, the first language model may be a statistical model or a neural network model, which is not limited in the embodiments of the present disclosure.

[0053] Step 202: in response to the input abnormal feedback information, obtain the actual text information corresponding to the audio information.

[0054] Step 202 has the same or similar function as the above-mentioned step 102, and its implementation method can be performed with reference to the above-mentioned embodiment, which will not be described in detail here.

[0055] Step 203: Determine an abnormal word sequence in the first recognition result based on the abnormal feedback information.

[0056] Step 204: determine the actual word sequence in the actual text information corresponding to the abnormal word sequence.

[0057] In some embodiments of the present disclosure, the abnormal feedback information is obtained by identifying abnormal user operations. By determining that the current first recognition result is abnormal, by identifying the abnormal feedback information fed back when the user performs abnormal operations, an abnormal word sequence in the first recognition result and an actual word sequence corresponding to the abnormal word sequence are determined.

[0058] In an optional embodiment of the present disclosure, bad case identification can be performed on user operations. For example, the audio information input during the navigation process is "wo yao qu gao xing yi dian", and the first recognition result obtained by the first language model recognition is "I want to go to be happier". When abnormal feedback information such as the user's evaluation of "wrong recognition", a second click on the recognition result of the navigation address, multiple interactions with the same content, etc. is identified, the actual text information "I want to go to Gao Xing Yi Dian" corresponding to "wo yao qu gao xing yi dian" can be obtained. At the same time, through the abnormal feedback information, it is determined that the abnormal word sequence in the current first recognition result is "happier" and the actual word sequence "Gao Xing Yi Dian" corresponding to the abnormal word sequence is determined.

[0059] Step 205, based on the actual word sequence, the first recognition result is corrected to obtain a second recognition result, and the correction processing includes at least one of the first correction processing, the second correction processing, the third correction processing, and the fourth correction processing.

[0060] In some embodiments of the present disclosure, the specific processing methods of the first correction process, the second correction process, the third correction process, and the fourth correction process are different, and the impacts on the recognition results are also different. The correction process for the first recognition result can be performed by one correction method, or multiple correction methods can be combined to correct the first result, which is not limited in the present disclosure.

[0061] In some embodiments of the present disclosure, based on the actual word sequence, the first correction process for the first recognition result includes: replacing the abnormal word sequence with the actual word sequence to correct the first recognition result.

[0062] In an alternative embodiment of the present disclosure, the first correction process can directly replace the error result, i.e., the abnormal word sequence. For example, when "Navigate to Gaoxingyidian" is recognized as "Navigate to Happy a bit", the abnormal word sequence in the first recognition result can be replaced as a whole. Replace "Happy a bit" in "Navigate to Happy a bit" with "Gaoxingyidian" to achieve the overall replacement of the first recognition result and directly replace the first recognition result with the correct "Navigate to Gaoxingyidian".

[0063] It should be noted that the abnormal word sequence will be completely deleted after the first correction process.

[0064] In some embodiments of the present disclosure, based on the actual word sequence, the second correction process for the first recognition result includes: using the first weighting coefficient to increase the probability value of the actual word sequence in the candidate recognition result to correct the first recognition result.

[0065] The first weighting coefficient is preset, and the specific value is set according to the actual situation, which is not limited in the embodiments of the present disclosure. By using the first weighting coefficient, the probability value of the actual word sequence can be increased, thereby increasing the probability value of the recognition result containing the actual word sequence in the candidate recognition result.

[0066] In an alternative embodiment of the present disclosure, the overall score of the actual word sequence (correct result) can be enhanced, that is, the probability value of the word sequence is increased. For example, the probability value of "Gaoxingyidian" is increased, thereby increasing the overall probability value of the candidate recognition result containing "Gaoxingyidian" in the candidate recognition result.

[0067] In some embodiments of the present disclosure, performing a third correction processing on the first recognition result based on the actual word sequence includes: using personalized information to construct a second language model, the personalized information includes the actual word sequence; recognizing audio information through the second language model to obtain the probability of the third recognition result; determining a second weighting coefficient and a third weighting coefficient; using the second weighting coefficient and the third weighting coefficient, weightedly fusing the probability of the third recognition result with the probability of the first recognition result to correct the first recognition result, the probability of the first recognition result being obtained through the first language model.

[0068] The personalized information is not limited to the actual word sequence, but may also include user preference information.

[0069] In an optional embodiment of the present disclosure, the actual word sequence is used for modeling, and the second speech model (i.e., language model sub-model) is used for modeling to improve the probability of user preference information (common words, custom terms, etc.), geographic location (nearby address names, home and company names, etc.), media names (user playlists, recent popular songs, etc.) and other user personalized information. The second language model is used to identify the audio information and obtain a relevance score P. 副 , that is, the probability of the third recognition result; at the same time, the first language model (i.e., the main language model) provides a general score P 主 , that is, the probability of the first recognition result. Select a suitable coefficient and fuse the probability of the first recognition result with the probability of the third recognition result to adjust the probability of the first recognition result and perform badcase repair.

[0070] The weighted fusion method is shown in Formula 1, including but not limited to linear weighting.

[0071] Total score = k 主 P 主 +k 副 P 副 Formula 1

[0072] k 主 is the third weighting coefficient, P 主 is the probability of the first recognition result, k 副 is the second weighting coefficient, P 副 is the probability of the third recognition result.

[0073] It should be noted that the second language model includes but is not limited to a statistical model and a neural network model.

[0074] In some embodiments of the present disclosure, the first language model is a statistical model, and based on the actual word sequence, the fourth correction processing of the first recognition result includes: adding the actual word sequence to the candidate recognition result of the first language model, or deleting the abnormal word sequence from the candidate recognition result of the first language model, so as to correct the first recognition result.

[0075] In an optional embodiment of the present disclosure, when the first language model is a statistical model, the candidate recognition results of the first language model are adjusted. For example, when the first language model is an n-gram language model, the actual word sequence can be directly added to the candidate recognition results of the first language model, or the abnormal word sequence can be deleted from the candidate recognition results of the first language model to improve the probability of correct content appearing.

[0076] Step 206: In response to the abnormal feedback information corresponding to the second recognition result, the second recognition result is corrected.

[0077] In some embodiments of the present disclosure, when a second recognition result (i.e., a corrected first recognition result) is obtained, if the user inputs abnormal feedback information again, the second recognition result is corrected again according to the current abnormal feedback information, and the repair strength can be continuously adjusted by changing the correction processing method.

[0078] In an optional embodiment of the present disclosure, if the user continues to make multiple attempts and complain about the recognition result after the fourth correction processing is used for repair, the current recognition result, i.e., the second recognition result, is considered to be still a bad case, and the correction method is changed to use the third correction processing method for enhancement. If the obtained recognition result is still identified as a bad case, the coefficient in the third correction processing can be further adjusted to enhance the correction effect of the third correction processing.

[0079] If the above method still fails to achieve a recognition result that meets the user's needs, that is, the recognition effect is not good, the second correction processing method can be directly used to strongly intervene in the probability value of the actual word sequence, that is, directly increase the probability value of the actual word sequence. Similarly, if the recognition result still does not meet the needs, the abnormal word sequence in the recognition result can be directly replaced and modified by the first correction processing method.

[0080] It is understandable that different correction processing methods in the present disclosure have different correction effects. In the case of correcting the second recognition result, the correction processing method can be adaptively selected according to actual needs and is not limited in the present embodiment. With the continuous iteration of weights and strategies, the performance of the language model can also be improved according to user habits, giving a more accurate recognition result when identifying the target.

[0081] In summary, according to the embodiments of the present disclosure, speech recognition is performed on the input audio information to obtain a first recognition result, and in response to the input abnormal feedback information, the actual text information corresponding to the audio information is obtained, and based on the abnormal feedback information, the abnormal word sequence in the first recognition result is determined, and the actual word sequence in the actual text information corresponding to the abnormal word sequence is determined. Based on the actual word sequence, the first recognition result is corrected to obtain a second recognition result, and the correction processing includes at least one of the first correction processing, the second correction processing, the third correction processing, and the fourth correction processing. In response to the abnormal feedback information corresponding to the second recognition result, the second recognition result is corrected to achieve correction of the erroneous results of speech recognition, solve the problem that homophones are difficult to distinguish during speech recognition, and improve the accuracy of speech recognition. At the same time, the erroneous results can be adaptively corrected through a variety of correction methods, and the user preferences and user habits of the user are utilized to provide a new speech recognition system that fits the user's usage habits. After continuous iteration, the performance of the language model will increase with the user's use, thereby improving the correction efficiency of the language recognition results and the accuracy of speech recognition.

[0082] Corresponding to the above-mentioned method for correcting a speech recognition result, the present disclosure also proposes a device for correcting a speech recognition result. Figure 3 A schematic diagram of the structure of a speech recognition result correction device 300 provided in an embodiment of the present disclosure.

[0083] like Figure 3 As shown, the device is configured in a vehicle, and the device includes:

[0084] The recognition unit 310 is used to perform speech recognition on the input audio information to obtain a first recognition result;

[0085] The acquisition unit 320 is used to acquire actual text information corresponding to the audio information in response to the input abnormal feedback information;

[0086] The processing unit 330 is used to correct the first recognition result according to the actual text information to obtain a second recognition result, where the second recognition result is the actual recognition result corresponding to the audio information.

[0087] In some embodiments, the recognition unit 310 is used to: recognize audio information through a first language model, and obtain candidate recognition results corresponding to the audio information, where the candidate recognition results include the first recognition result.

[0088] In some embodiments, the processing unit 330 is used to: determine an abnormal word sequence in the first recognition result based on abnormal feedback information; determine an actual word sequence in the actual text information corresponding to the abnormal word sequence; and perform correction processing on the first recognition result based on the actual word sequence, the correction processing including at least one of a first correction processing, a second correction processing, a third correction processing, and a fourth correction processing.

[0089] In some embodiments, the processing unit 330 is used to replace the abnormal word sequence with the actual word sequence to correct the first recognition result.

[0090] In some embodiments, the processing unit 330 is used to: utilize the first weighting coefficient to increase the probability value of the actual word sequence in the candidate recognition result to correct the first recognition result.

[0091] In some embodiments, the processing unit 330 is used to: construct a second language model using personalized information, the personalized information including an actual word sequence; recognize audio information through the second language model to obtain a probability of a third recognition result; determine a second weighting coefficient and a third weighting coefficient; and use the second weighting coefficient and the third weighting coefficient to weightedly fuse the probability of the third recognition result with the probability of the first recognition result to correct the first recognition result, the probability of the first recognition result being obtained through the first language model.

[0092] In some embodiments, the first language model is a statistical model, and the processing unit 330 is used to add the actual word sequence to the candidate recognition result of the first language model, or delete the abnormal word sequence from the candidate recognition result of the first language model to correct the first recognition result.

[0093] In some embodiments, the processing unit 330 is further used to: in response to abnormal feedback information corresponding to the second recognition result, perform correction processing on the second recognition result.

[0094] According to an embodiment of the present disclosure, a first recognition result is obtained by performing speech recognition on input audio information; in response to input abnormal feedback information, actual text information corresponding to the audio information is obtained; the first recognition result is corrected through the actual text information to obtain a second recognition result, which is the actual recognition result corresponding to the audio information, thereby correcting erroneous results of speech recognition, solving the problem of homophones being difficult to distinguish during speech recognition, and improving the accuracy of speech recognition.

[0095] It should be noted that since the device embodiment of the present disclosure corresponds to the above-mentioned method embodiment, the above-mentioned explanation of the method embodiment is also applicable to the device of the present embodiment, and the principle is the same. For details not disclosed in the device embodiment, reference can be made to the above-mentioned method embodiment, and they will not be repeated in this disclosure.

[0096] According to an embodiment of the present disclosure, the present disclosure provides a vehicle, the aforementioned speech recognition result correction method can be applied to the vehicle, and the aforementioned speech recognition result correction device can be configured in the vehicle.

[0097] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0098] Figure 4 A schematic block diagram of an example electronic device 400 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0099] like Figure 4 As shown, the device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 402 or a computer program loaded from a storage unit 408 to a RAM (Random Access Memory) 403. In the RAM 403, various programs and data required for the operation of the device 400 can also be stored. The computing unit 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404. An I / O (Input / Output) interface 405 is also connected to the bus 404.

[0100] A number of components in the device 400 are connected to the I / O interface 405, including: an input unit 406, such as a keyboard, a mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a disk, an optical disk, etc.; and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the device 400 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0101] The computing unit 401 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, CPU (Central Processing Unit), GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as a method for correcting speech recognition results. For example, in some embodiments, the method for correcting speech recognition results may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to execute the aforementioned method for correcting the speech recognition result in any other appropriate manner (for example, by means of firmware).

[0102] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor that may be a dedicated or general-purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0103] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0104] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0105] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0106] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0107] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0108] It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0109] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0110] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for correcting speech recognition results, characterized in that: The method comprises: Performing speech recognition on the input audio information to obtain a first recognition result; In response to the input abnormal feedback information, obtaining actual text information corresponding to the audio information; The first speech recognition result is corrected using the actual text information to obtain a second recognition result, where the second recognition result is the actual recognition result corresponding to the audio information.

2. The method according to claim 1, characterized in that The performing speech recognition on the input audio information to obtain a first recognition result includes: The audio information is recognized by using a first language model, and candidate recognition results corresponding to the audio information are obtained, where the candidate recognition results include a first recognition result.

3. The method according to claim 2, characterized in that The correcting the first recognition result by using the actual text information includes: Based on the abnormal feedback information, determining an abnormal word sequence in the first recognition result; Determine an actual word sequence in the actual text information corresponding to the abnormal word sequence; Based on the actual word sequence, the first recognition result is corrected, and the corrected process includes at least one of a first corrected process, a second corrected process, a third corrected process, and a fourth corrected process.

4. The method according to claim 3, characterized in that Based on the actual word sequence, performing a first correction process on the first recognition result includes: The abnormal word sequence is replaced by the actual word sequence to correct the first recognition result.

5. The method according to claim 3, characterized in that: Based on the actual word sequence, performing a second correction process on the first recognition result includes: The first weighting coefficient is used to increase the probability value of the actual word sequence in the candidate recognition result to correct the first recognition result.

6. The method according to claim 3, characterized in that Based on the actual word sequence, performing a third correction process on the first recognition result includes: constructing a second language model using personalized information, wherein the personalized information includes actual word sequences; Recognize the audio information by using the second language model to obtain a probability of a third recognition result; determining a second weighting coefficient and a third weighting coefficient; The probability of the third recognition result is weighted and fused with the probability of the first recognition result using the second weighting coefficient and the third weighting coefficient to correct the first recognition result, wherein the probability of the first recognition result is obtained through the first language model.

7. The method according to claim 3, characterized in that The first language model is a statistical model, and based on the actual word sequence, performing a fourth correction process on the first recognition result includes: adding the actual word sequence to the candidate recognition results of the first language model; Alternatively, the abnormal word sequence is deleted from the candidate recognition result of the first language model to correct the first recognition result.

8. The method according to any one of claims 1 to 7, characterized in that The method further comprises: In response to the abnormal feedback information corresponding to the second recognition result, the second recognition result is corrected.

9. A speech recognition result correction device, characterized in that: The device comprises: A recognition unit, configured to perform speech recognition on the input audio information to obtain a first recognition result; An acquisition unit, configured to acquire actual text information corresponding to the audio information in response to the input abnormal feedback information; The processing unit is used to correct the first recognition result according to the actual text information to obtain a second recognition result, where the second recognition result is the actual recognition result corresponding to the audio information.

10. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-8.

12. A vehicle, characterized in that: It includes the speech recognition result correction device as described in claim 9 or the electronic device as described in claim 10.

Images