Speech recognition method and device, computer device and storage medium

Abstract

The application relates to a speech recognition method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: performing speech recognition on audio data to obtain first text information and the occurrence time of each character in the first text information; performing filtering processing on the audio data based on the frequency of the main frequency in the audio data to obtain filtered audio; performing speech recognition on the filtered audio to obtain second text information and the text frequency corresponding to the second text information; filtering the first text information based on the occurrence time and the text frequency to obtain third text information; and performing semantic recognition processing on the third text information to obtain a speech recognition result of the audio data. The method can improve the accuracy of speech recognition.

Description

Technical Field

[0001] This application relates to the field of information recognition technology, and in particular to a speech recognition method, apparatus, computer equipment, storage medium and computer product. Background Technology

[0002] With the development of communication technology and the widespread use of smartphones, various online communication tools and software communication products have gradually become one of the main tools for public communication. Currently, some communication applications already have the technology to convert voice information into text.

[0003] However, existing technologies only focus on the semantic recognition of speech information, which makes the text information recognized from speech more susceptible to interference, resulting in a low accuracy rate of speech recognition. Summary of the Invention

[0004] Therefore, it is necessary to provide a speech recognition method, device, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of speech recognition in order to address the above-mentioned technical problems.

[0005] In a first aspect, this application provides a speech recognition method, the method comprising:

[0006] Perform speech recognition on the audio data to obtain the first text information and the occurrence time of each character in the first text information;

[0007] Based on the frequency of the main frequency in the audio data, the audio data is filtered to obtain filtered audio;

[0008] The filtered audio is subjected to speech recognition to obtain second text information and the text frequency corresponding to the second text information;

[0009] Based on the occurrence time and the frequency of the text, the first text information is filtered to obtain the third text information;

[0010] The third text information is subjected to semantic recognition processing to obtain the speech recognition result of the audio data.

[0011] In one embodiment, the step of performing speech recognition on the audio data to obtain first text information and the occurrence time of each character in the first text information includes:

[0012] The trained speech recognition model is used to perform speech recognition on audio data containing speech, and the first text information consisting of multiple characters is obtained.

[0013] The occurrence time of each character is determined based on the pronunciation time of each character in the audio data in the first text information.

[0014] In one embodiment, filtering the audio data based on the frequency of the dominant frequency in the audio data to obtain filtered audio includes:

[0015] The audio data is transformed using Fourier transform to obtain multiple amplitudes belonging to the audio data, and the target amplitude with the largest amplitude value is obtained among the amplitudes.

[0016] Based on the frequency of the target amplitude, determine the main frequency of the audio data;

[0017] Based on the main frequency, the audio data is filtered to obtain the filtered audio corresponding to the audio data.

[0018] In one embodiment, the step of performing speech recognition on the filtered audio to obtain second text information and the text frequency corresponding to the second text information includes:

[0019] The filtered audio is then subjected to speech recognition using the trained speech recognition model to obtain second text information composed of multiple characters.

[0020] Based on the pronunciation time of each character in the second text information in the filtered audio, the pronunciation interval between the characters is determined;

[0021] Based on the average value of the pronunciation intervals, the frequency of the character corresponding to the second character information is determined.

[0022] In one embodiment, filtering the first text information based on the occurrence time and the text frequency to obtain the third text information includes:

[0023] Mark the time of occurrence of each character in the first text information on the time axis, and superimpose an identifier used to represent the frequency of the character to obtain the initial superposition result;

[0024] Adjust the phase of the identifier in the initial superposition result to obtain the degree of overlap between the phase-adjusted identifier and the occurrence time of each of the characters;

[0025] Obtain the target overlap degree that meets the target condition from the overlap degree, and extract the text corresponding to each overlap time under the target overlap degree to obtain the third text information.

[0026] In one embodiment, the step of performing semantic recognition processing on the third text information to obtain the speech recognition result of the audio data includes:

[0027] Configure a semantic rule information base;

[0028] The semantic rule information base is used to identify and process the third text information to obtain text information that does not match the semantic rule information base;

[0029] The text information is converted and processed to obtain target text information that matches the semantic rule information database;

[0030] Based on the target text information, the speech recognition result of the audio data is determined.

[0031] Secondly, this application provides a speech recognition device, the device comprising:

[0032] The first speech recognition module is used to perform speech recognition on audio data to obtain first text information and the occurrence time of each character in the first text information.

[0033] The filtered audio acquisition module is used to filter the audio data based on the frequency of the main frequency in the audio data to obtain filtered audio;

[0034] The second speech recognition module is used to perform speech recognition on the filtered audio to obtain second text information and the text frequency corresponding to the second text information.

[0035] An information filtering module is used to filter the first text information based on the occurrence time and the text frequency to obtain the third text information;

[0036] The recognition result acquisition module is used to perform semantic recognition processing on the third text information to obtain the speech recognition result of the audio data.

[0037] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0038] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0039] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0040] The aforementioned speech recognition methods, devices, computer equipment, storage media, and computer program products filter the first text information based on the first text information using second text information and text frequencies obtained from the filtering audio based on the dominant frequency. Since the second text information is not directly used in terms of text content, information in the audio data can be avoided from being missed. By filtering the first text information using text frequencies and occurrence times, text that does not conform to the text frequencies can be removed. Finally, semantic recognition is performed based on the third text information, improving the accuracy of speech-to-text conversion and thus achieving accurate speech recognition of audio data. Attached Figure Description

[0041] Figure 1 This is a diagram illustrating the application environment of a speech recognition method in one embodiment;

[0042] Figure 2 This is a flowchart illustrating a speech recognition method in one embodiment;

[0043] Figure 3 This is a flowchart illustrating the speech recognition method in another embodiment;

[0044] Figure 4 This is a structural block diagram of a speech recognition device in one embodiment;

[0045] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] The speech recognition method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Server 104 performs speech recognition processing on the acquired audio data to obtain each pronounced character in the audio data, i.e., the first character information, and also the pronunciation time of each character in the audio data, i.e., the time of appearance. Server 104 filters the acquired audio data according to the dominant frequency in the audio data, thereby obtaining filtered audio with interference removed. Server 104 performs speech recognition processing on the filtered audio with interference removed to obtain the second character information appearing in the filtered audio, and also obtains the character frequency corresponding to the second character information. Server 104 performs filtering processing on the obtained first character information based on the time of appearance of each character in the first character information and the character frequency corresponding to the second character information, obtaining the third character information. Server 104 performs semantic recognition processing on the third character information according to semantic recognition processing principles to obtain the speech recognition result of the audio data. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0048] In one embodiment, such as Figure 2 As shown, a speech recognition method is provided, which is applied to... Figure 1 Taking the server in the example, the following steps are included:

[0049] Step 202: Perform speech recognition on the audio data to obtain the first text information and the occurrence time of each character in the first text information.

[0050] Audio data refers to electronic files containing speech. The main formats of audio data include AVI (AudioVideo Interleaved), AAC (Advanced Audio Coding), and MP3 (Moving Picture Experts Group Audio Layer 3).

[0051] Speech recognition is accomplished using speech recognition models. A speech recognition model is a model generated by training a machine learning model; it is an end-to-end model.

[0052] The first text information refers to each character identified by the speech recognition model from the audio data. The moment of occurrence refers to the moment when each character is pronounced in the audio data.

[0053] Optionally, the server uses a trained speech recognition model to process the audio data containing speech, obtaining first text information composed of the recognized characters and the pronunciation time of each recognized character in the audio data.

[0054] Step 204: Based on the frequency of the main frequency in the audio data, filter the audio data to obtain filtered audio.

[0055] The dominant frequency refers to the frequency of the position with the highest amplitude in the acquired audio data. For example, in a segment of audio data, if point A has the highest amplitude, then the frequency of point A is called the dominant frequency.

[0056] Filtering is the process of removing specific frequency bands from a signal, and it is an important measure to suppress and prevent interference. Filtering methods mainly include low-pass filtering, high-pass filtering, wavelet filtering, adaptive filtering, and matched filtering.

[0057] Optionally, the server determines the dominant frequency of the audio data based on the amplitude at each point in the audio data. Based on the dominant frequency, the server uses filtering techniques to remove all signals from the audio data except for the dominant frequency and its surrounding areas to prevent interference.

[0058] Step 206: Perform speech recognition on the filtered audio to obtain the second text information and the text frequency corresponding to the second text information.

[0059] The second text information refers to each character identified by the speech recognition model from the filtered audio. The character frequency corresponding to the second text information is obtained by calculating the pronunciation interval between every two adjacent characters in the second text information in the filtered audio.

[0060] Optionally, the server uses a trained speech recognition model to process the filtered audio obtained after filtering out interference factors, and obtains second text information composed of multiple characters. By calculating the pronunciation interval between each pair of adjacent characters in the second text information and the filtered audio, the text frequency corresponding to the second text information is obtained.

[0061] Step 208: Based on the time of occurrence and the frequency of the characters, filter the first text information to obtain the third text information.

[0062] The server filters the first text information obtained from the audio data based on the occurrence time of each character in the first text information and the character frequency based on the pronunciation interval between each pair of adjacent characters in the second text information, thereby obtaining the third text information composed of multiple characters.

[0063] Step 210: Perform semantic recognition processing on the third text information to obtain the speech recognition result of the audio data.

[0064] Semantic recognition is an important component of natural language processing technology. The core of semantic recognition is not only to understand the meaning of words in text, but also to understand the meaning of words in sentences and paragraphs. This means that semantic recognition technically requires semantic analysis and ambiguity elimination at the text, vocabulary, syntax, lexical and paragraph levels, as well as corresponding meaning reconstruction, in order to achieve the purpose of recognition itself.

[0065] Optionally, the server uses semantic recognition technology in natural language processing to recognize and process the third text information in order to understand the meaning of the text words in the third text information, convert the speech into text, and obtain the speech recognition result of the audio data.

[0066] In the aforementioned speech recognition method, based on the first text information, the second text information and text frequency obtained from the filtered audio based on the dominant frequency are used to filter the first text information. Since the second text information is not directly used in terms of text content, information in the audio data can be avoided from being missed. By filtering the first text information using text frequency and occurrence time, text that does not conform to the text frequency can be removed. Finally, semantic recognition is performed based on the third text information, improving the accuracy of speech-to-text conversion and thus achieving accurate speech recognition of audio data.

[0067] In one embodiment, performing speech recognition on audio data to obtain first text information and the occurrence time of each character in the first text information includes:

[0068] The trained speech recognition model is used to perform speech recognition on audio data containing speech, resulting in first text information composed of multiple characters.

[0069] Based on the pronunciation time of each character in the audio data in the first text information, determine the time of appearance of each character.

[0070] Among them, the speech recognition model is a model generated by training a machine learning model, and it is an end-to-end model. Using an end-to-end model in speech recognition can reduce the preprocessing and feature processing of the raw speech data, directly converting the speech data into text data, reducing the errors introduced by processing operations, and improving the efficiency and accuracy of speech recognition.

[0071] Optionally, the server uses a trained speech recognition model to perform speech recognition on the acquired audio data, directly converting the speech data into text data to obtain the first text information composed of multiple characters. The server then determines the occurrence time of each character in the audio data based on the pronunciation time of each recognized character.

[0072] In this embodiment, by using an end-to-end speech recognition model, the preprocessing and feature processing of audio data can be reduced, thereby reducing errors introduced by processing operations and improving the efficiency and accuracy of speech recognition.

[0073] In one embodiment, filtering the audio data based on the dominant frequency in the audio data to obtain filtered audio includes:

[0074] The audio data is transformed using Fourier transform to obtain multiple amplitudes belonging to the audio data, and the target amplitude with the largest amplitude value is obtained.

[0075] The dominant frequency of the audio data is determined based on the frequency of the target amplitude.

[0076] Based on the main frequency, the audio data is filtered to obtain the filtered audio corresponding to the audio data.

[0077] The Fourier transform can represent a function that satisfies certain conditions as a linear combination of trigonometric functions or their integrals. In different research fields, the Fourier transform has various variants, such as the continuous Fourier transform and the discrete Fourier transform.

[0078] The target amplitude represents the amplitude at which the physical quantity of vibration reaches its maximum value in the audio data. The dominant frequency refers to the frequency at which the physical quantity of vibration reaches its maximum value in the acquired audio data after the information in the time domain has been converted into information in the frequency domain using Fourier transform.

[0079] Optionally, the server uses Fourier transform to convert the audio data from time-domain information to frequency-domain information, thereby obtaining the physical quantities of vibration in the audio data. The server selects the physical quantity with the largest vibration from the obtained vibration physical quantities as the target amplitude. Based on the frequency corresponding to the target amplitude, the server determines the dominant frequency of the audio data. The server uses a filter processor to filter out all interference factors except for the dominant frequency and its surrounding areas, obtaining the filtered audio.

[0080] In this embodiment, by performing a Fourier transform on the audio data, the dominant frequency of the audio data can be accurately determined, thereby filtering the audio data to obtain accurately filtered audio.

[0081] In one embodiment, speech recognition is performed on the filtered audio to obtain second text information and the text frequency corresponding to the second text information, including:

[0082] The trained speech recognition model is used to perform speech recognition on the filtered audio to obtain second text information composed of multiple characters.

[0083] The pronunciation interval between characters is determined based on the pronunciation time of each character in the second text information in the filtered audio.

[0084] The frequency of the second text information is determined based on the average value of the pronunciation interval.

[0085] The pronunciation interval refers to the duration of the interval between the pronunciations of two adjacent characters. Character frequency refers to the frequency of a character's occurrence. For example, if the sound of one character is produced within 10 seconds, the character frequency is 0.1 characters per second.

[0086] Optionally, the server uses a trained speech recognition model to perform speech recognition on the filtered audio, converting the speech data into text data to obtain second text information composed of multiple characters. The server calculates the pronunciation interval between any two adjacent characters based on the pronunciation time of each recognized character in the filtered audio. The server then calculates the average pronunciation interval of each pair of adjacent characters in the second text information to obtain the character frequency corresponding to the second text information.

[0087] In this embodiment, by using an end-to-end speech recognition model, the preprocessing and feature processing of filtered audio can be reduced, thereby reducing errors introduced by processing operations and improving the efficiency and accuracy of speech recognition.

[0088] In one embodiment, such as Figure 3 As shown, the step of filtering the first text information based on the occurrence time and the text frequency to obtain the third text information includes:

[0089] Step 302: Mark the time of appearance of each character in the first text information on the time axis, and superimpose the identifier used to represent the frequency of the characters to obtain the initial superposition result.

[0090] Among them, the identifier for the frequency of a character is an identifier used to represent the numerical value of the frequency of a character, and the identifier can be in the form of a graphic.

[0091] Optionally, the server marks the pronunciation time of each character in the first text information on the time axis, and overlays the character frequency, which is the average pronunciation interval of every two adjacent characters in the filtered audio, on the time axis in a graphical form to obtain the initial overlay result.

[0092] Step 304: Adjust the phase of the identifier in the initial superposition result to obtain the overlap between the phase-adjusted identifier and the occurrence time of each character.

[0093] Here, phase is actually an angle, corresponding to the angle of vector rotation.

[0094] Optionally, the server adjusts the angle of the identifier represented by the graphic in the initial superposition result so that the adjusted identifier graphic coincides with the pronunciation time of the first character on the time coordinate axis. Based on the result of the coincidence, the server determines the degree of coincidence after the phase adjustment.

[0095] Step 306: Obtain the target overlap degree that meets the target condition in the overlap degree, and extract the text corresponding to each overlap time under the target overlap degree to obtain the third text information.

[0096] The target condition refers to the highest degree of overlap. The target degree of overlap is the degree of overlap between the identifier of the character frequency and the mark on the time axis of the pronunciation time of each character in the first text information.

[0097] Optionally, after multiple phase adjustments, the service obtains multiple degrees of overlap. From these multiple degrees of overlap, the server selects the highest degree of overlap after phase adjustment, which is the target degree of overlap, and extracts the text at each overlap time under the highest degree of overlap to obtain the third text information.

[0098] In this embodiment, by filtering the first text information by text frequency, texts that match the text frequency can be obtained, and other text interference elements can be eliminated, thereby improving the accuracy of speech recognition.

[0099] In one embodiment, semantic recognition processing is performed on the third text information to obtain the speech recognition result of the audio data, including:

[0100] Configure the semantic rule information base.

[0101] The semantic rule information database is used to identify and process the third-party text information, resulting in text information that does not match the semantic rule information database.

[0102] The text information is converted and processed to obtain the target text information that matches the semantic rule information database.

[0103] Based on the target text information, determine the speech recognition result of the audio data.

[0104] The semantic rule information base is pre-generated based on Chinese grammar and is used to detect whether the positional relationship between each word segment and its adjacent words in the text is correct. The conversion process refers to converting mismatched text information into homophones or near-homophones until it matches the semantic rule information base.

[0105] Optionally, the server generates a semantic rule information database based on Chinese grammar, and uses this database to detect each character in the third-party text information. Characters that do not match the semantic rule information database are converted into homophones or near-homophones until they match the database. Based on the obtained target text information, the server obtains the speech recognition result of the audio data.

[0106] In this embodiment, by configuring a semantic rule information base, the recognized text can be further calibrated based on the third text information, thereby obtaining accurate speech recognition results.

[0107] This application also provides an application scenario in which the above-described speech recognition method is applied. Specifically, the speech recognition method is applied in this scenario as follows: The obtained audio data is processed using a trained speech recognition model to directly convert the speech signal in the audio data into text, obtaining first text information composed of multiple characters. The occurrence time of each character is determined based on its pronunciation time in the audio data. Fourier transform is used to convert the audio data from the time domain to the frequency domain, obtaining the amplitude of the audio data, and the frequency of the amplitude with the largest value is determined as the dominant frequency. A filter is used to filter out interference factors other than the dominant frequency and its surrounding areas, obtaining filtered audio. The speech recognition model is used to perform speech recognition on the filtered audio, obtaining second text information composed of multiple characters. The pronunciation interval between every two adjacent characters is obtained based on the pronunciation time of each character in the second text information in the filtered audio. The character frequency is obtained based on the average pronunciation interval between every two adjacent characters in the second text information. The occurrence time of each character in the first text information is marked on a time axis, and identifiers representing character frequencies are superimposed on the same axis. The phase of the character frequency identifiers is continuously adjusted to maximize the overlap between the identifiers and the occurrence times. The characters at each overlap time in the state of maximum overlap are recorded to obtain the third text information. A semantic rule information database is generated based on Chinese grammar to detect each character in the third text information. Characters that do not match the semantic rule information database are converted into homophones or near-homophones until they match the semantic rule information database to obtain the target text information. Based on the obtained target text information, the speech recognition result of the audio data is obtained.

[0108] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0109] Based on the same inventive concept, this application also provides a speech recognition device for implementing the speech recognition method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more speech recognition device embodiments provided below can be found in the limitations of the speech recognition method described above, and will not be repeated here.

[0110] In one embodiment, such as Figure 4 As shown, a voice recognition device is provided, comprising:

[0111] The first speech recognition module 402 is used to perform speech recognition on audio data to obtain first text information and the occurrence time of each character in the first text information.

[0112] The audio filtering acquisition module 404 is used to filter the audio data based on the frequency of the main frequency in the audio data to obtain filtered audio.

[0113] The second speech recognition module 406 is used to perform speech recognition on the filtered audio to obtain the second text information and the text frequency corresponding to the second text information.

[0114] The information filtering module 408 is used to filter the first text information based on the time of occurrence and the frequency of the text to obtain the third text information.

[0115] The recognition result acquisition module 410 is used to perform semantic recognition processing on the third text information to obtain the speech recognition result of the audio data.

[0116] In one embodiment, the first speech recognition module includes:

[0117] The first text information acquisition unit is used to perform speech recognition on audio data containing speech using a trained speech recognition model to obtain first text information composed of multiple characters.

[0118] The occurrence time determination unit is used to determine the occurrence time of each character based on the pronunciation time of each character in the audio data in the first text information.

[0119] In one embodiment, the filtered audio acquisition module includes:

[0120] The target amplitude determination unit is used to transform the audio data using Fourier transform to obtain multiple amplitudes belonging to the audio data, and to obtain the target amplitude with the largest amplitude value among the amplitudes.

[0121] The main frequency determination unit is used to determine the main frequency of the audio data based on the frequency of the target amplitude.

[0122] The filtered audio acquisition unit is used to filter audio data based on the main frequency to obtain the filtered audio corresponding to the audio data.

[0123] In one embodiment, the second speech recognition module includes: a second text information determination unit, used to perform speech recognition on the filtered audio using a trained speech recognition model to obtain second text information composed of multiple characters.

[0124] The pronunciation interval determination unit is used to determine the pronunciation interval between characters based on the pronunciation time of each character in the second text information in the filtered audio.

[0125] The character frequency determination unit is used to determine the character frequency of the second character information corresponding to 5 based on the average value of the pronunciation interval.

[0126] In one embodiment, the information filtering module includes:

[0127] The initial superposition result acquisition unit is used to mark the occurrence time of each character in the first text information on the time coordinate axis and superimpose the identifier used to represent the frequency of the characters to obtain the initial superposition result.

[0128] The phase adjustment unit is used to adjust the phase of the identifier in the initial superposition result, so as to obtain the overlap between the identifier with the phase of 0 and the occurrence time of each character.

[0129] The third text information acquisition unit is used to acquire the target overlap degree that meets the target conditions in the overlap degree, and extract the text corresponding to each overlap time under the target overlap degree to obtain the third text information.

[0130] In one embodiment, the recognition result acquisition module includes:

[0131] The rule configuration unit is used to configure the semantic rule information base.

[0132] 5. Information recognition and processing unit, used to recognize and process third-party text information using a semantic rule information database, to obtain text information that does not match the semantic rule information database.

[0133] The target text information acquisition unit is used to convert and process text information to obtain target text information that matches the semantic rule information database.

[0134] The recognition result acquisition unit is used to determine the speech recognition result of the audio data based on the target text information.

[0135] Each module in the aforementioned speech recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0136] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores audio data, first text information, the occurrence time of each character in the first text information, its frequency, filtered audio, second text information, character frequencies, third text information, and speech recognition result data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a speech recognition method.

[0137] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0138] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0139] Speech recognition is performed on the audio data to obtain the first text information and the occurrence time of each character in the first text information. Based on the dominant frequency in the audio data, the audio data is filtered to obtain filtered audio. Speech recognition is performed on the filtered audio to obtain the second text information and the corresponding character frequencies of the second text information. Based on the occurrence time and character frequencies, the first text information is filtered to obtain the third text information. Semantic recognition processing is performed on the third text information to obtain the speech recognition result of the audio data.

[0140] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0141] The trained speech recognition model is used to perform speech recognition on audio data containing speech, resulting in a first text information composed of multiple characters. The occurrence time of each character is determined based on its pronunciation time in the audio data.

[0142] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0143] The audio data is transformed using Fourier transform to obtain multiple amplitudes belonging to the audio data, and the target amplitude with the largest amplitude value is obtained. Based on the frequency of the target amplitude, the dominant frequency of the audio data is determined. Based on the dominant frequency, the audio data is filtered to obtain the filtered audio data.

[0144] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0145] The trained speech recognition model is used to perform speech recognition on the filtered audio, resulting in second text information composed of multiple characters. Based on the pronunciation time of each character in the second text information within the filtered audio, the pronunciation interval between characters is determined. Based on the average pronunciation interval, the frequency of the corresponding character in the second text information is determined.

[0146] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0147] The occurrence time of each character in the first text information is marked on a time axis, and an identifier representing the character frequency is superimposed to obtain an initial superposition result. The phase of the identifier in the initial superposition result is adjusted to obtain the overlap degree between the phase-adjusted identifier and the occurrence time of each character. The target overlap degree that meets the target condition is obtained, and the character corresponding to each overlap time under the target overlap degree is extracted to obtain the third text information.

[0148] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0149] Configure a semantic rule information base. Use the semantic rule information base to recognize and process the third-party text information, obtaining text information that does not match the semantic rule information base. Transform the text information to obtain target text information that matches the semantic rule information base. Based on the target text information, determine the speech recognition result of the audio data.

[0150] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0151] Speech recognition is performed on the audio data to obtain the first text information and the occurrence time of each character in the first text information. Based on the dominant frequency in the audio data, the audio data is filtered to obtain filtered audio. Speech recognition is performed on the filtered audio to obtain the second text information and the corresponding character frequencies of the second text information. Based on the occurrence time and character frequencies, the first text information is filtered to obtain the third text information. Semantic recognition processing is performed on the third text information to obtain the speech recognition result of the audio data.

[0152] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0153] The trained speech recognition model is used to perform speech recognition on audio data containing speech, resulting in a first text information composed of multiple characters. The occurrence time of each character is determined based on its pronunciation time in the audio data.

[0154] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0155] The audio data is transformed using Fourier transform to obtain multiple amplitudes belonging to the audio data, and the target amplitude with the largest amplitude value is obtained. Based on the frequency of the target amplitude, the dominant frequency of the audio data is determined. Based on the dominant frequency, the audio data is filtered to obtain the filtered audio data.

[0156] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0157] The trained speech recognition model is used to perform speech recognition on the filtered audio, resulting in second text information composed of multiple characters. Based on the pronunciation time of each character in the second text information within the filtered audio, the pronunciation interval between characters is determined. Based on the average pronunciation interval, the frequency of the corresponding character in the second text information is determined.

[0158] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0159] The occurrence time of each character in the first text information is marked on a time axis, and an identifier representing the character frequency is superimposed to obtain an initial superposition result. The phase of the identifier in the initial superposition result is adjusted to obtain the overlap degree between the phase-adjusted identifier and the occurrence time of each character. The target overlap degree that meets the target condition is obtained, and the character corresponding to each overlap time under the target overlap degree is extracted to obtain the third text information.

[0160] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0161] Configure a semantic rule information base. Use the semantic rule information base to recognize and process the third-party text information, obtaining text information that does not match the semantic rule information base. Transform the text information to obtain target text information that matches the semantic rule information base. Based on the target text information, determine the speech recognition result of the audio data.

[0162] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0163] Speech recognition is performed on the audio data to obtain the first text information and the occurrence time of each character in the first text information. Based on the dominant frequency in the audio data, the audio data is filtered to obtain filtered audio. Speech recognition is performed on the filtered audio to obtain the second text information and the corresponding character frequencies of the second text information. Based on the occurrence time and character frequencies, the first text information is filtered to obtain the third text information. Semantic recognition processing is performed on the third text information to obtain the speech recognition result of the audio data.

[0164] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0165] The trained speech recognition model is used to perform speech recognition on audio data containing speech, resulting in a first text information composed of multiple characters. The occurrence time of each character is determined based on its pronunciation time in the audio data.

[0166] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0167] The audio data is transformed using Fourier transform to obtain multiple amplitudes belonging to the audio data, and the target amplitude with the largest amplitude value is obtained. Based on the frequency of the target amplitude, the dominant frequency of the audio data is determined. Based on the dominant frequency, the audio data is filtered to obtain the filtered audio data.

[0168] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0169] The trained speech recognition model is used to perform speech recognition on the filtered audio, resulting in second text information composed of multiple characters. Based on the pronunciation time of each character in the second text information within the filtered audio, the pronunciation interval between characters is determined. Based on the average pronunciation interval, the frequency of the corresponding character in the second text information is determined.

[0170] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0171] The occurrence time of each character in the first text information is marked on a time axis, and an identifier representing the character frequency is superimposed to obtain an initial superposition result. The phase of the identifier in the initial superposition result is adjusted to obtain the overlap degree between the phase-adjusted identifier and the occurrence time of each character. The target overlap degree that meets the target condition is obtained, and the character corresponding to each overlap time under the target overlap degree is extracted to obtain the third text information.

[0172] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0173] Configure a semantic rule information base. Use the semantic rule information base to recognize and process the third-party text information, obtaining text information that does not match the semantic rule information base. Transform the text information to obtain target text information that matches the semantic rule information base. Based on the target text information, determine the speech recognition result of the audio data.

[0174] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0175] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0176] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0177] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A voice recognition method, characterized by, The method comprises: performing speech recognition on the audio data to obtain first text information and respective occurrence time of each text in the first text information; performing filtering processing on the audio data based on the frequency of the main frequency in the audio data to obtain filtered audio; performing speech recognition on the filtered audio to obtain second text information and text frequency corresponding to the second text information; filtering the first text information based on the occurrence time and the text frequency to obtain third text information; performing semantic recognition processing on the third text information to obtain the speech recognition result of the audio data; wherein filtering the first text information based on the occurrence time and the text frequency to obtain third text information comprises: marking the occurrence time of each text in the first text information in a time coordinate axis, and superimposing identifiers for representing the text frequency to obtain an initial superimposition result; adjusting the phase of the identifiers in the initial superimposition result to obtain the coincidence degree of the adjusted phase of the identifiers and the occurrence time of each text; obtaining target coincidence degrees that meet a target condition in the coincidence degrees, and extracting the text corresponding to each coincidence time under the target coincidence degrees to obtain the third text information.

2. The method of claim 1, wherein, The speech recognition on the audio data to obtain first text information and respective occurrence time of each text in the first text information comprises: performing speech recognition on the audio data containing speech using a trained speech recognition model to obtain first text information composed of multiple texts; determining the occurrence time of each text based on the pronunciation time of each text in the audio data.

3. The method of claim 1, wherein, The filtering processing on the audio data based on the frequency of the main frequency in the audio data to obtain filtered audio comprises: performing transformation processing on the audio data using Fourier transform to obtain multiple amplitudes belonging to the audio data, and obtaining a target amplitude with the maximum amplitude in the amplitudes; determining the main frequency frequency of the audio data based on the frequency of the target amplitude; performing filtering processing on the audio data based on the main frequency frequency to obtain filtered audio corresponding to the audio data.

4. The method of claim 1, wherein, The speech recognition on the filtered audio to obtain second text information and text frequency corresponding to the second text information comprises: performing speech recognition on the filtered audio using a trained speech recognition model to obtain second text information composed of multiple texts; determining the pronunciation interval between the texts based on the pronunciation time of each text in the filtered audio; determining the text frequency corresponding to the second text information based on the average value of the pronunciation interval.

5. The method of claim 1, wherein, The semantic recognition processing on the third text information to obtain the speech recognition result of the audio data comprises: configuring a semantic rule information base; performing recognition processing on the third text information using the semantic rule information base to obtain text information that does not match the semantic rule information base; The text information is converted to obtain target text information matched with the semantic rule information base; Based on the target text information, a speech recognition result of the audio data is determined.

6. The method of claim 3, wherein, The method further comprises: The audio data is converted from time domain information to frequency domain information using Fourier transform to obtain a physical quantity of vibration of the audio data; The physical quantity with the largest vibration is selected as a target amplitude from the physical quantity of vibration.

7. A speech recognition apparatus characterized by comprising: The device comprises: A first speech recognition module is configured to perform speech recognition on audio data to obtain first text information and an occurrence time of each character in the first text information; A filtered audio acquisition module is configured to perform filtering processing on the audio data based on a frequency of a main frequency in the audio data to obtain filtered audio; A second speech recognition module is configured to perform speech recognition on the filtered audio to obtain second text information and a character frequency corresponding to the second text information; An information filtering module is configured to filter the first text information based on the occurrence time and the character frequency to obtain third text information; An identification result acquisition module is configured to perform semantic recognition processing on the third text information to obtain a speech recognition result of the audio data; The information filtering module comprises: an initial superposition result acquisition unit configured to mark the occurrence time of each character in the first text information in a time coordinate axis and superimpose an identifier representing the character frequency to obtain an initial superposition result; a phase adjustment unit configured to adjust the phase of the identifier in the initial superposition result to obtain a coincidence degree of the adjusted phase of the identifier and the occurrence time of each character; and a third text information acquisition unit configured to acquire a target coincidence degree that meets a target condition from the coincidence degree and extract a character corresponding to each coincidence time under the target coincidence degree to obtain the third text information.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images