Method, device and equipment for recognizing voice containing hot words and storage device

Abstract

The invention provides a hot word-containing speech recognition method and device, equipment and a storage device. The method comprises the following steps: acquiring a to-be-recognized speech; performing preliminary voice recognition on the to-be-recognized voice to obtain a preliminary recognition text; matching each preliminary recognition text with a hot word database, and correcting the confidence coefficient of the preliminary recognition text when the preliminary recognition text contains hot words; and determining a target recognition text according to the confidence of all the preliminary recognition texts. Through the method, the change of a language environment can be dynamically adapted, new popular vocabularies can be captured in time, even for newest hot words which are not specially trained, accurate recognition can be performed through context information and a hot word prefix tree, different application scenes and language environments can be dynamically adapted, and the precision and reliability of a recognition result are improved.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to a speech recognition method, device, equipment and storage device containing hot words. Background Art

[0002] Speech recognition technology has been widely used in the fields of intelligent assistants, voice control, translation equipment, etc. The current speech recognition system mainly relies on the combination of acoustic models and language models, and converts speech into text through feature extraction and semantic analysis of speech signals.

[0003] However, with the development of the Internet, new hot words and buzzwords emerge in an endless stream, and traditional speech recognition systems have difficulty recognizing these new words in a timely manner. As a result, in actual use, new words used by users may not be correctly recognized, thus affecting the user experience.

[0004] In summary, the problems existing in the prior art need to be solved urgently. Summary of the invention

[0005] The present invention provides a speech recognition method, device, equipment and storage device containing hot words, which are used to solve the defects in the prior art and realize speech recognition containing hot words.

[0006] The present invention provides a speech recognition method including hot words, comprising:

[0007] Get the speech to be recognized;

[0008] Performing preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text;

[0009] Matching each of the preliminary recognized texts with a hot word database, and when the preliminary recognized text contains a hot word, correcting the confidence of the preliminary recognized text, wherein the confidence is used to determine the priority of the preliminary recognized text;

[0010] The target recognition text is determined based on the confidence of all preliminary recognition texts.

[0011] According to a speech recognition method containing hot words provided by the present invention, the step of matching each of the preliminary recognized texts with a hot word database and correcting the confidence of the preliminary recognized text when the preliminary recognized text contains a hot word specifically includes:

[0012] Obtaining the initial confidence and confidence amplification factor of the hot word;

[0013] Determining a rationality score of the hot word according to context information of the hot word;

[0014] The target confidence of the hot word is determined according to the initial confidence, the confidence amplification factor and the rationality score.

[0015] According to a speech recognition method including hot words provided by the present invention, after the step of determining the target confidence of the hot words according to the initial confidence, the confidence amplification factor and the rationality score, the method further includes:

[0016] When the target confidence is lower than a preset confidence threshold, a replacement word is determined according to the hot word and the context information;

[0017] The hot words in the initial recognition text are replaced with the replacement words.

[0018] According to a speech recognition method containing hot words provided by the present invention, the initial confidence is calculated in the following manner:

[0019]

[0020] Where v is a vocabulary sequence, v′ is all other vocabulary sequences different from v, x is an audio signal, P(x|v) is the probability output by the acoustic model, and P(v) is the probability output by the language model;

[0021] The language model is used to estimate the likelihood of a given word sequence, and the acoustic model outputs a probability distribution over all possible words.

[0022] According to a method for speech recognition containing hot words provided by the present invention, before the step of obtaining the speech to be recognized, the method further includes:

[0023] Build a hot word database;

[0024] The hot word data is acquired regularly and saved in the hot word database.

[0025] According to a speech recognition method containing hot words provided by the present invention, the step of matching each of the preliminary recognized texts with a hot word database and correcting the confidence of the preliminary recognized text when the preliminary recognized text contains a hot word specifically includes:

[0026] Obtaining the initial confidence and confidence amplification factor of the hot word;

[0027] Obtain video data of the speech to be recognized;

[0028] Performing visual recognition on the video data to obtain visual recognition data;

[0029] A target confidence of the hot word is determined according to the initial confidence, the confidence amplification factor and the visual recognition data.

[0030] According to a speech recognition method including hot words provided by the present invention, the visual recognition includes environment recognition and lip shape recognition and action recognition of the speaker.

[0031] The present invention also provides a speech recognition device including hot words, comprising:

[0032] A speech acquisition module, used to acquire the speech to be recognized;

[0033] A speech recognition module, used for performing preliminary speech recognition on the speech to be recognized to obtain a preliminary recognition text;

[0034] A confidence correction module, used for matching each of the preliminary recognized texts with a hot word database, and when the preliminary recognized text contains a hot word, correcting the confidence of the preliminary recognized text, wherein the confidence is used to determine the priority of the preliminary recognized text;

[0035] The text determination module is used to determine the target recognition text according to the confidence of all the preliminary recognition texts.

[0036] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a speech recognition method containing hot words as described in any one of the above methods is implemented.

[0037] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for speech recognition containing hot words.

[0038] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for speech recognition containing hot words.

[0039] The speech recognition method, device, equipment and storage device containing hot words provided by the present invention obtain the speech to be recognized; perform preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text; match each of the preliminary recognition texts with the hot word database, and when the preliminary recognition text contains a hot word, correct the confidence of the preliminary recognition text; determine the target recognition text according to the confidence of all preliminary recognition texts. This method can dynamically adapt to changes in the language environment and capture new popular words in a timely manner. Even for the latest hot words that have not been specially trained, they can be accurately recognized through context information and hot word prefix trees, and can dynamically adapt to different application scenarios and language environments, thereby improving the accuracy and reliability of the recognition results. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 It is a flow chart of a speech recognition method including hot words provided by the present invention;

[0042] Figure 2 It is a structural schematic diagram of a speech recognition device including hot words provided by the present invention;

[0043] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] In order to solve the problems in the prior art, the present invention proposes a speech recognition method containing hot words to realize speech recognition containing hot words. The speech recognition method containing hot words is described below. Figure 1 As shown, including but not limited to the following steps:

[0046] Step 110: Acquire the speech to be recognized.

[0047] In this step, the voice signal to be recognized is acquired through a voice acquisition device. The voice acquisition device can be a microphone, a recording device, a smart phone or other device with a voice acquisition function. The collected voice signal can be a real-time voice input or a pre-recorded audio file. The voice signal can include audio forms such as natural language conversations, voice commands or speeches. During the acquisition process, the voice signal will be pre-processed, such as noise elimination, voice enhancement, etc., to improve the accuracy of subsequent recognition.

[0048] Step 120: Perform preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text.

[0049] When performing preliminary speech recognition, the preprocessed speech signal is input into the speech recognition module. The speech recognition module can use a combination of an acoustic model and a language model based on a deep neural network. It first analyzes the spectral characteristics of the audio signal and then predicts the possible text sequence through the model. The acoustic model is responsible for converting the audio signal into specific vocabulary units (such as phonemes or words), while the language model combines these units according to the context and outputs possible text results.

[0050] The output of this step is one or more preliminary recognition texts, each of which corresponds to a certain confidence value, which is used to indicate the degree of trust the speech recognition module has in the text result. The preliminary recognition text can be a list of multiple candidate results, and the system will assign a corresponding confidence to each candidate result.

[0051] Step 130: Match each of the preliminary recognized texts with a hot word database. When the preliminary recognized text contains a hot word, correct the confidence of the preliminary recognized text. The confidence is used to determine the priority of the preliminary recognized text.

[0052] After the initial recognition text is generated, it is matched with a pre-built hot word database. The hot word database can consist of a set of domain-specific keywords, proper nouns, or user-defined terms to improve the recognition accuracy related to specific scenarios. The matching process can be achieved through simple string matching algorithms or other context-based semantic analysis algorithms.

[0053] If a hot word is detected in the preliminary recognition text, the confidence of the preliminary recognition text will be corrected based on the following information:

[0054] Hot word initial confidence: In the hot word database, each hot word is usually set with an initial confidence to indicate the importance of the hot word in a specific context.

[0055] Contextual rationality score: The system further evaluates the possibility of the hot word appearing by analyzing the contextual rationality of the text in which the hot word appears. If the hot word appears within a reasonable semantic range, its confidence will increase; otherwise, it may decrease.

[0056] Confidence amplification factor: If the rationality of a hot word is high, the system will introduce an amplification factor to further improve the confidence of the text related to the hot word.

[0057] Through the above adjustments, the system can dynamically modify the confidence of text results containing hot words, thereby improving their priority among candidate results.

[0058] Step 140: Determine the target recognition text according to the confidence of all the preliminary recognition texts.

[0059] After all the preliminary recognition texts have been matched with hot words and confidence correction, the text with the highest confidence is selected as the target recognition text according to the final confidence value of each text. In some cases, if the confidence of multiple texts is similar, further context analysis, external knowledge base query or user feedback mechanism can be used to assist in the final determination of the target recognition text.

[0060] The target recognition text is the output of this step, which will eventually be provided to the user as the result of speech recognition or used for subsequent processing.

[0061] In this embodiment, the construction of the hot word database can be based on multiple approaches, for example:

[0062] Manual addition: Users can manually define and add hot words based on their specific application scenarios.

[0063] Dynamic update: The system can automatically obtain new hot words from the Internet, terminology libraries in specific fields, or user interactions at regular intervals, and update them to the hot word database.

[0064] The speech recognition method containing hot words provided by the present invention obtains the speech to be recognized; performs preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text; matches each of the preliminary recognition texts with a hot word database, and when the preliminary recognition text contains a hot word, corrects the confidence of the preliminary recognition text; and determines the target recognition text according to the confidence of all preliminary recognition texts. This method can dynamically adapt to changes in the language environment and capture new popular words in a timely manner. Even for the latest hot words that have not been specially trained, it can be accurately recognized through context information and hot word prefix trees, and can dynamically adapt to different application scenarios and language environments, thereby improving the accuracy and reliability of the recognition results.

[0065] As a further optional embodiment, the step of matching each of the preliminary recognized texts with a hot word database and correcting the confidence of the preliminary recognized text when the preliminary recognized text contains a hot word specifically includes:

[0066] Obtaining the initial confidence and confidence amplification factor of the hot word;

[0067] Determining a rationality score of the hot word according to context information of the hot word;

[0068] The target confidence of the hot word is determined according to the initial confidence, the confidence amplification factor and the rationality score.

[0069] In this embodiment, it is first necessary to obtain the initial confidence and confidence amplification factor of the hot word. In the hot word database, each hot word usually has an initial confidence value, which reflects the importance or frequency of the hot word in a specific context. The initial confidence can be determined by a domain expert or a machine learning model based on historical data, and is usually generated when the hot word database is built. At the same time, a confidence amplification factor is obtained, which is used to dynamically adjust the confidence according to the rationality of the context in subsequent steps. The confidence amplification factor is a parameter used to adjust the initial confidence according to the accuracy of the hot word context.

[0070] Next, the rationality score of the hot word is determined based on the context information of the hot word. Context information is an important basis for judging the rationality of hot words in the recognized text. The system calculates the rationality score of the hot word by analyzing the context around the hot word (such as adjacent words, grammatical structures, voice and intonation, etc.). For example, when a word in a specific field appears in a conversation in that field, the rationality score may be higher; if the hot word appears in a conversation context that is unrelated to it, the rationality score will be lower. In addition, the system can also combine external knowledge bases (such as industry terms, encyclopedia information, etc.) to further confirm whether the appearance of the hot word in the context conforms to conventional semantic rules, thereby deriving a rationality score.

[0071] Finally, the target confidence of the hot word is determined according to the initial confidence, the confidence amplification factor and the rationality score. Specifically, target confidence = initial confidence × (1 + confidence amplification factor × rationality score). The initial confidence provides the basic confidence, and the confidence amplification factor adjusts the confidence improvement according to the rationality score. If the rationality score is higher, the improvement of the target confidence will be more obvious; otherwise, the target confidence will not be greatly improved, and may even decrease.

[0072] In this way, the system can dynamically adjust the confidence of the hot word according to its rationality and importance in the context, and combine it with the confidence of other recognized texts to improve the accuracy and reliability of the final recognition result. This method is particularly suitable for multi-domain and multi-scenario speech recognition systems, so that hot words in specific scenarios can be better recognized and displayed first.

[0073] As a further optional embodiment, after the step of determining the target confidence of the hot word according to the initial confidence, the confidence amplification factor and the rationality score, the method further includes:

[0074] When the target confidence is lower than a preset confidence threshold, a replacement word is determined according to the hot word and the context information;

[0075] The hot words in the initial recognition text are replaced with the replacement words.

[0076] In this embodiment, it is first determined whether the target confidence is lower than a preset confidence threshold. A confidence threshold is preset to determine whether the confidence of a hot word in the recognized text is high enough. If the target confidence is lower than the threshold, it means that the recognition reliability of the hot word is low, and there may be misrecognition or inaccuracy.

[0077] This threshold can be dynamically adjusted according to different application scenarios. For example, for professional terms in the medical or legal fields, the confidence threshold can be set higher, while for some non-critical words, the threshold can be appropriately relaxed.

[0078] Next, a replacement word is determined based on the hot word and context information. When the target confidence is lower than a preset threshold, a possible replacement word needs to be determined by analyzing the hot word and its context information. The replacement word can be determined in the following ways:

[0079] Context analysis: Based on the words that appear before and after the hot word, infer the words that are more semantically appropriate. For example, in a specific sentence structure, a more grammatically appropriate word can be used as a replacement word.

[0080] Hot word similarity database: Use the vocabulary database related to the hot word to find words that are close in meaning to the original hot word but have higher recognition reliability. This vocabulary database can be extracted from the domain term library or trained and updated through a machine learning model.

[0081] Historical recognition data: You can use historical recognition results and user feedback to learn which words are more likely to appear in that context and use this data to determine the most appropriate replacement words.

[0082] Finally, the hot words in the preliminary recognition text are replaced with replacement words. Once the replacement words are determined, the hot words in the preliminary recognition text are replaced with the replacement words. This operation can help improve the accuracy of the final recognition results, especially in the target recognition text generation process, making the text more in line with user expectations.

[0083] The replaced text is evaluated again to ensure it fits the overall context and language structure.

[0084] For example, in a speech recognition scenario in the medical field, the hot word "cell" identified in the initial recognition text may have a low confidence level. After context analysis, it is found that the more reasonable word in this context should be "cell membrane", so "cell" is replaced with "cell membrane", thereby improving the accuracy of the final text.

[0085] This hot word replacement mechanism can further improve the accuracy of speech recognition. Especially in cases of low confidence, it can intelligently provide reasonable alternatives to avoid the adverse effects of misrecognition.

[0086] As a further optional embodiment, the initial confidence is calculated in the following manner:

[0087]

[0088] Where v is a vocabulary sequence, v′ is all other vocabulary sequences different from v, x is an audio signal, P(x|v) is the probability output by the acoustic model, and P(v) is the probability output by the language model;

[0089] The language model is used to estimate the likelihood of a given word sequence, and the acoustic model outputs a probability distribution over all possible words.

[0090] As a further optional embodiment, the initial confidence is calculated by the following formula:

[0091]

[0092] Where: v is a vocabulary sequence, representing a specific word or phrase; v′ is all other possible vocabulary sequences different from v; x is an audio signal, representing the audio input to be recognized; P(x|v) is the probability output by the acoustic model, representing the probability of the audio signal x appearing when the vocabulary sequence is v; P(v) is the probability output by the language model, representing the prior probability of the vocabulary sequence v appearing.

[0093] Through the above formula, the system can combine the output of the acoustic model and the language model to calculate the confidence C(v) corresponding to each word sequence v. Specifically, the acoustic model evaluates each possible word through the audio signal, while the language model estimates the probability of the word appearing in natural language based on the given context and word sequence.

[0094] The calculation process includes the following steps:

[0095] Acoustic model evaluation: The acoustic model analyzes the audio signal and outputs the probability of each word, that is, P(x|v).

[0096] Language model evaluation: The language model evaluates the probability of each word appearing, i.e., P(v), based on the context and prior knowledge.

[0097] Normalization: Multiply the acoustic and linguistic probabilities of all possible word sequences and normalize them to get the final confidence of each word.

[0098] This calculation method generates more accurate confidence results by integrating information from both audio signals and language structures for subsequent recognition decisions. This method is particularly suitable for scenarios involving complex contexts and multiple interpretation possibilities, balancing the audio feature recognition of the acoustic model and the contextual understanding of the language model in the confidence calculation.

[0099] As a further optional embodiment, before the step of obtaining the speech to be recognized, the method further includes:

[0100] Build a hot word database;

[0101] The hot word data is acquired regularly and saved in the hot word database.

[0102] In this embodiment, a hot word database specifically used to store new network words is constructed according to predetermined application scenarios and field requirements. Hot words specifically refer to new words that frequently appear and spread rapidly in the network environment in recent years. These words may be related to popular culture, hot events or network trends, and are commonly seen in social media, news reports and forum discussions.

[0103] The construction process of the hot word database can be achieved in many ways, including crawling the web and automatically extracting new words from the web. For example, using web crawler technology to automatically extract hot words from social media, forums and other platforms, and filtering out representative new words through natural language processing technology.

[0104] In order to ensure the timeliness and dynamism of the hot word database, the system will regularly update the network new word data. This can be achieved through an automated Internet crawling mechanism or with the help of an external hot word API interface. The update cycle can be set according to the application scenario (such as daily, weekly or monthly) to ensure that the latest network new words can be added to the database in a timely manner.

[0105] Automatic crawling: The system can regularly obtain the latest Internet hot words from popular social media platforms, news portals, video comments, etc. By analyzing the keywords and tags of these platforms, it can identify new words that appear frequently and have a spreading trend in the current network.

[0106] User feedback updates: When users find that some new Internet words have not been correctly recognized, they can submit these words through the feedback mechanism. The system will record and add them to the hot word library after review to ensure that they are correctly processed in subsequent recognition.

[0107] In order to optimize performance, hot word data can be saved in an incremental update manner to avoid large-scale reconstruction of the entire database. Only newly added hot words are processed and updated to reduce system load.

[0108] As a further optional embodiment, the step of matching each of the preliminary recognized texts with a hot word database and correcting the confidence of the preliminary recognized text when the preliminary recognized text contains a hot word specifically includes:

[0109] Obtaining the initial confidence and confidence amplification factor of the hot word;

[0110] Obtain video data of the speech to be recognized;

[0111] Performing visual recognition on the video data to obtain visual recognition data;

[0112] A target confidence of the hot word is determined according to the initial confidence, the confidence amplification factor and the visual recognition data.

[0113] In this embodiment, an initial confidence value is assigned to each hot word according to its historical performance in the speech recognition model or its record in the hot word database, and the initial confidence value reflects the possibility of the hot word being correctly recognized in a specific speech environment.

[0114] In addition, a confidence amplification factor is assigned based on the application scenario and the importance of the hot word. This amplification factor is used to adjust the confidence of the hot word during the recognition process to ensure more flexible and accurate processing of speech content with hot words in actual scenarios. For example, hot words related to popular events can have their priority increased by a larger confidence amplification factor.

[0115] While processing speech recognition, the system also acquires video data that is synchronized with the speech signal. This data can come from the user's device (such as a camera) or other associated monitoring equipment. The purpose of acquiring video data is to use multimodal information to enhance the accuracy of speech recognition, especially when the pronunciation is unclear or the speech signal is interfered with by noise.

[0116] Next, the video data is analyzed to extract visual features related to speech, mainly including the speaker's lip shape recognition and action recognition. By identifying the speaker's mouth movements (such as changes in lip shape) and facial expressions and body movements during speech, the system can assist in inferring the content of the pronunciation, thereby providing additional information for speech recognition.

[0117] In addition, environmental recognition can also be performed to determine the background information in the speech scene (such as the presence or absence of noise, the spatial environment in which the speaker is located, etc.). This information helps to further adjust the confidence of the recognized text.

[0118] Combined with visual recognition information extracted from video data, the system can more accurately determine the target confidence of hot words in speech.

[0119] First, by analyzing the matching degree between the speaker's mouth shape and the pronunciation of the hot word, the confidence of the hot word can be increased or decreased. If the pronunciation and mouth shape match well, the system will increase the confidence of the hot word, otherwise it will decrease.

[0120] Then, the speaker’s action recognition and environment recognition data are combined to further confirm the rationality of the hot word. For example, if the speaker mentions a hot word during a specific action (such as pointing to an object), the confidence of the hot word can be further improved.

[0121] Finally, the target confidence of the hot word is calculated comprehensively based on the initial confidence, confidence amplification factor and visual recognition data to ensure that the priority of the hot word in the recognition results is reasonably adjusted.

[0122] As a further optional embodiment, the visual recognition includes environment recognition and lip shape recognition and action recognition of the speaker.

[0123] Environmental Identification:

[0124] By analyzing the environmental information in the video, the background, scene and ambient sound of the speaker can be identified. For example, if the speaker is in a specific scene (such as a meeting or event venue), the system can further confirm whether the hot words in the speech content are in context by identifying relevant environmental elements (such as landmarks, background objects, etc.).

[0125] Lip recognition:

[0126] Through the mouth movements and lip shape changes in the video data, the system can identify the speaker's pronunciation when speaking, and match this information with the pronunciation pattern of the hot word. If the speaker's lip shape is identified to be highly consistent with a hot word, the system will increase the confidence of the hot word.

[0127] Action Recognition:

[0128] It can also identify the speaker's body movements and gestures to help infer the hot words they mentioned. For example, when mentioning a hot word, the speaker may point to a related object or make a specific action. These action information can be used to help determine the correctness of the hot word and adjust its confidence.

[0129] The following is a description of the hot word speech recognition device provided by the present invention. Figure 2 As shown, the speech recognition device for hot words described below and the speech recognition method for hot words described above can be referred to each other.

[0130] A hot word speech recognition device, comprising:

[0131] The speech acquisition module 210 is used to acquire the speech to be recognized;

[0132] The speech recognition module 220 is used to perform preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text;

[0133] A confidence correction module 230, used for matching each of the preliminary recognized texts with a hot word database, and when the preliminary recognized text contains a hot word, correcting the confidence of the preliminary recognized text, wherein the confidence is used to determine the priority of the preliminary recognized text;

[0134] The text determination module 240 is used to determine the target recognition text according to the confidence of all the preliminary recognition texts.

[0135] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communication interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute the speech recognition method containing hot words, and the method includes:

[0136] Get the speech to be recognized;

[0137] Performing preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text;

[0138] Matching each of the preliminary recognized texts with a hot word database, and when the preliminary recognized text contains a hot word, correcting the confidence of the preliminary recognized text, wherein the confidence is used to determine the priority of the preliminary recognized text;

[0139] The target recognition text is determined based on the confidence of all preliminary recognition texts.

[0140] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM, Read-On l yMemory), a random access memory (RAM, Random Access Memory), a disk or an optical disk.

[0141] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the speech recognition method containing hot words provided by the above methods, the method includes:

[0142] Get the speech to be recognized;

[0143] Performing preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text;

[0144] Matching each of the preliminary recognized texts with a hot word database, and when the preliminary recognized text contains a hot word, correcting the confidence of the preliminary recognized text, wherein the confidence is used to determine the priority of the preliminary recognized text;

[0145] The target recognition text is determined based on the confidence of all preliminary recognition texts.

[0146] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the speech recognition method containing hot words provided by the above methods, the method comprising:

[0147] Get the speech to be recognized;

[0148] Performing preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text;

[0149] Matching each of the preliminary recognized texts with a hot word database, and when the preliminary recognized text contains a hot word, correcting the confidence of the preliminary recognized text, wherein the confidence is used to determine the priority of the preliminary recognized text;

[0150] The target recognition text is determined based on the confidence of all preliminary recognition texts.

[0151] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0152] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A speech recognition method containing hot words, characterized in that: include: Get the speech to be recognized; Performing preliminary speech recognition on the speech to be recognized to obtain preliminary recognition text; Matching each of the preliminary recognized texts with a hot word database, and when the preliminary recognized text contains a hot word, correcting the confidence of the preliminary recognized text, wherein the confidence is used to determine the priority of the preliminary recognized text; The target recognition text is determined based on the confidence of all preliminary recognition texts.

2. The method for speech recognition containing hot words according to claim 1, characterized in that: The step of matching each of the preliminary recognized texts with the hot word database and correcting the confidence of the preliminary recognized text when the preliminary recognized text contains a hot word specifically includes: Obtaining the initial confidence and confidence amplification factor of the hot word; Determining a rationality score of the hot word according to context information of the hot word; The target confidence of the hot word is determined according to the initial confidence, the confidence amplification factor and the rationality score.

3. The method for speech recognition containing hot words according to claim 2, characterized in that: After the step of determining the target confidence of the hot word according to the initial confidence, the confidence amplification factor and the rationality score, the method further includes: When the target confidence is lower than a preset confidence threshold, a replacement word is determined according to the hot word and the context information; The hot words in the initial recognition text are replaced with the replacement words.

4. The method for speech recognition containing hot words according to claim 2, characterized in that: The initial confidence is calculated in the following way: Where v is a vocabulary sequence, v′ is all other vocabulary sequences different from v, x is an audio signal, P(x|v) is the probability output by the acoustic model, and P(v) is the probability output by the language model; The language model is used to estimate the likelihood of a given word sequence, and the acoustic model outputs a probability distribution over all possible words.

5. The method for speech recognition containing hot words according to claim 1, characterized in that: Before the step of obtaining the speech to be recognized, the method further includes: Build a hot word database; The hot word data is acquired regularly and saved in the hot word database.

6. The method for speech recognition containing hot words according to claim 1, characterized in that: The step of matching each of the preliminary recognized texts with the hot word database and correcting the confidence of the preliminary recognized text when the preliminary recognized text contains a hot word specifically includes: Obtaining the initial confidence and confidence amplification factor of the hot word; Obtain video data of the speech to be recognized; Performing visual recognition on the video data to obtain visual recognition data; A target confidence of the hot word is determined according to the initial confidence, the confidence amplification factor and the visual recognition data.

7. The method for speech recognition containing hot words according to claim 6, characterized in that: The visual recognition includes environment recognition and speaker's lip shape recognition and action recognition.

8. A speech recognition device including hot words, characterized in that: include: A speech acquisition module, used to acquire the speech to be recognized; A speech recognition module, used for performing preliminary speech recognition on the speech to be recognized to obtain a preliminary recognition text; A confidence correction module, used for matching each of the preliminary recognized texts with a hot word database, and when the preliminary recognized text contains a hot word, correcting the confidence of the preliminary recognized text, wherein the confidence is used to determine the priority of the preliminary recognized text; The text determination module is used to determine the target recognition text according to the confidence of all the preliminary recognition texts.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the speech recognition method containing hot words as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for speech recognition containing hot words as claimed in any one of claims 1 to 7 is implemented.

Images