Word adding method and device, equipment, storage medium, program product

Abstract

The application provides a word adding method and device, equipment, a storage medium, and a program product, which include the following steps: determining a pronunciation similar word of a target word in different language environments; determining a target word similarity according to a pronunciation similarity between voice data corresponding to the target word and a to-be-compared word in different language environments; if the target word similarity meets a preset similarity condition, obtaining a voice recognition result of the voice data corresponding to the target word in different noise environments; determining a comprehensive recognition reliability of the target word according to the target word similarity and the voice recognition result; and if the preset reliability condition is met, adding the target word to a preset hot word library. The risk of incorrect recognition of the target word can be determined according to the similarity between the target word and the to-be-compared word and the voice recognition result of the target word, and then it can be determined whether the target word is added to the preset hot word library, thereby improving the accuracy of voice recognition of the hot word in the preset word library after addition.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and includes, but is not limited to, a word addition method, apparatus, device, storage medium, and program product. Background Technology

[0002] Speech recognition is the process of identifying input speech data to obtain the corresponding text content. In the fields of speech recognition and related technologies, hot words typically refer to words in the speech-text content obtained through speech recognition that possess specific semantics or can be received by a designated system and trigger corresponding operations. By constructing a lexicon containing multiple hot words, the correspondence between different hot words and interactive operations can be determined. Specifically, a designated system acquires and processes external speech data, identifies whether specified hot words exist, and then, based on the correspondence between the specified hot words and preset interactive operations, controls home appliances, vehicles, or other interactive objects, improving the convenience of user operation.

[0003] In related technologies, the impact of user pronunciation on hot word recognition is often ignored when constructing hot word databases. For example, similar pronunciations may correspond to different text content in different regional dialects. Therefore, in speech recognition, some words will be incorrectly identified as hot words in the database, triggering erroneous responses from the system. Consequently, both the accuracy of hot word recognition and the user experience may fail to meet requirements. Summary of the Invention

[0004] In view of this, the word addition method, apparatus, device, storage medium, and program product provided in this application can determine the comprehensive recognition reliability of the target word based on the similarity between the target word and words with similar pronunciations and existing hot words in a preset hot word library, as well as the result of speech recognition of the target word, and determine whether to add the target word to the preset hot word library, thereby improving the accuracy of speech recognition of hot words in the preset word library after addition. The word addition method, apparatus, device, storage medium, and program product provided in this application are implemented as follows:

[0005] The word addition method provided in this application includes:

[0006] Identify multiple words with similar pronunciations to the target word in different linguistic environments, including Mandarin environments and dialect environments of the target region;

[0007] Based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments, the target word similarity between the target word and the word to be compared is determined. The word to be compared includes the plurality of similar pronunciation words and the plurality of hot words contained in the preset hot word library.

[0008] When the similarity of the target words meets the preset similarity conditions, the speech data corresponding to the target words is processed for speech recognition in different noise environments to obtain the speech recognition results of the target words.

[0009] Based on the target word similarity and the speech recognition results, the overall recognition reliability of the target word is determined;

[0010] If the overall recognition reliability meets the preset reliability conditions, the target word is added to the preset hot word library.

[0011] In some embodiments, determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments includes:

[0012] According to a preset speech recognition algorithm, the pronunciation similarity between the spectral features of the speech data corresponding to the target word in different language environments and the spectral features of the speech data corresponding to the word to be compared is obtained;

[0013] Based on the frequency of occurrence of the word to be compared in the pre-stored historical text dataset, the frequency of use of the word to be compared is obtained. The historical text dataset includes text data corresponding to multiple previously acquired speech data.

[0014] The similarity between the target word and the word to be compared is determined based on the pronunciation similarity and the frequency of use.

[0015] In some embodiments, determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments includes:

[0016] Based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the multiple similar-sounding words in different language environments, the degree of misidentification of the target word is determined. The degree of misidentification is used to characterize the degree to which misidentification is easily caused due to the size of the pronunciation similarity.

[0017] If the degree of misidentification meets the preset misidentification conditions, the similarity between the target word and the word to be compared is determined based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the multiple hot words.

[0018] In some embodiments, determining the degree of misidentification of the target word based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the plurality of similarly pronounced words in different language environments includes:

[0019] Based on the frequency of occurrence of the multiple similar-sounding words in a pre-stored historical text dataset, the frequency of use of the multiple similar-sounding words is obtained. The historical text dataset includes text data corresponding to multiple previously acquired speech data.

[0020] If the number of the plurality of similar-sounding words is less than a preset first number, at least some of the similar-sounding words among the plurality of similar-sounding words are determined to be popular similar words based on the frequency of use of different similar-sounding words;

[0021] If the number of popular similar words is less than a preset second quantity requirement, based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the multiple pronunciation similar words in different language environments, the number of misidentified language environments with a pronunciation similarity greater than a preset threshold is obtained.

[0022] Based on the number of misidentified environments, the pronunciation similarity between the speech data corresponding to the target word in different language environments and the speech data corresponding to the multiple similar-sounding words, and the frequency of use of different similar-sounding words, the overall misidentification situation corresponding to the multiple similar-sounding words is determined.

[0023] The degree of misidentification of the target word is determined based on the comprehensive misidentification situation, the number of words with similar pronunciations, and the number of popular similar words.

[0024] In some embodiments, determining multiple phonetically similar words of the target word in different linguistic environments includes:

[0025] In the different language environments, the spectral features of the speech data corresponding to the target word are compared with the spectral features of the speech data corresponding to words in the preset word library to determine words with similar pronunciations that have a higher feature similarity than a preset similarity threshold.

[0026] In some embodiments, the step of performing speech recognition processing on the speech data corresponding to the target word in different noise environments to obtain the speech recognition result of the target word includes:

[0027] Speech recognition processing is performed on the speech data corresponding to the target words in different noise environments to obtain the recognized text data of the speech data corresponding to the target words.

[0028] The recognized text data obtained in different noisy environments is matched with the target words to obtain speech recognition results including the number of successful matches and the success rate of matching.

[0029] The word addition device provided in this application embodiment includes:

[0030] The similar word acquisition module is used to identify multiple pronunciation similar words of the target word in different linguistic environments, including Mandarin environment and target regional dialect environment;

[0031] The similarity assessment module is used to determine the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments. The word to be compared includes the plurality of pronunciation similar words and the plurality of hot words contained in the preset hot word library.

[0032] The reliability assessment module is used to perform speech recognition processing on the speech data corresponding to the target word in different noise environments when the similarity of the target word meets the preset similarity conditions, and obtain the speech recognition result of the target word; and determine the comprehensive recognition reliability of the target word based on the similarity of the target word and the speech recognition result.

[0033] An additional processing module is added to add the target word to the preset hot word library when the overall recognition reliability meets the preset reliability conditions.

[0034] The computer device provided in this application includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements the method described in this application.

[0035] The computer-readable storage medium provided in this application embodiment stores a computer program thereon, which, when executed by a processor, implements the method provided in this application embodiment.

[0036] The computer program product provided in this application includes a computer program that, when executed by a processor, implements the method described in this application.

[0037] The word addition method, apparatus, computer device, and computer-readable storage medium provided in this application identify multiple pronunciation-similar words for a target word in different linguistic environments, including Mandarin and target regional dialect environments. For different dialect environments, the pronunciation similarity between the target word and the speech data of a word to be compared, including pronunciation-similar words and hot words in a preset hot word library, is compared to determine the target word's similarity. The higher the target word's similarity, the more likely it is to be misidentified as a word to be compared during subsequent speech recognition, and therefore, it is unsuitable for addition to the preset hot word library. If the target word's similarity meets preset conditions, the method provided in this application further performs speech recognition processing on the target word's speech data under different noise environments to obtain the speech recognition results. Combining the target word's similarity and the speech recognition results, the comprehensive recognition reliability of the target word is determined. Only when the comprehensive recognition reliability meets preset reliability conditions is the target word added to the preset hot word library.

[0038] In this way, before adding the target word to the preset hot word library, the similarity between the target word and words with similar pronunciations and existing hot words in the preset hot word library is determined. Combined with the recognition results of speech recognition of the target word, the overall recognition reliability of the target word is judged, which reduces the probability of the target word being incorrectly recognized after being added to the preset hot word library and improves the accuracy of speech recognition of hot words in the preset word library after addition. Attached Figure Description

[0039] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application.

[0040] Figure 1 This is an application scenario diagram of the word addition method provided in the embodiments of this application;

[0041] Figure 2 A flowchart illustrating an implementation of the word addition method provided in this application embodiment;

[0042] Figure 3 A flowchart illustrating another implementation of the word addition method provided in this application embodiment;

[0043] Figure 4 A flowchart illustrating yet another implementation of the word addition method provided in this application embodiment;

[0044] Figure 5 A schematic diagram of a word addition device is provided for an embodiment of this application;

[0045] Figure 6This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the specific technical solutions of this application will be further described in detail below with reference to the accompanying drawings of the embodiments of this application. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0048] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0049] It should be noted that the terms "first, second, third" used in the embodiments of this application are used to distinguish similar or different objects and do not represent a specific order of objects. It can be understood that "first, second, third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0050] Speech recognition technology, also known as Automatic Speech Recognition (ASR), aims to convert the lexical content of human speech into computer-readable input. Natural Language Processing (NLP) is an important branch of speech recognition technology. Its goal is to build a bridge between machine language and human language to achieve human-computer interaction. Specifically, this involves a computer collecting and processing the user's speech to obtain corresponding text data, which can then be processed accordingly.

[0051] Currently, speech recognition and related technologies can accurately identify words in speech and have been widely applied in specific fields such as intelligent customer service, voice assistants, and smart home appliances. Users can interact with these systems or trigger specific operations through voice control. To achieve these specific functions, it is necessary to identify the words that trigger the corresponding operations; in related technical fields, these words are called hot words. For example, in intelligent customer service, a user might say, "I want to check my bill." In this example, "check" and "bill" are hot words. After recognizing these hot words, the system knows that the user's purpose is to check the bill and then executes the corresponding query operation. Similarly, in voice assistant applications, command words such as "open," "close," and "play" can also be set as hot words and stored in the corresponding system's hot word library. When the user speaks, speech recognition technology retrieves the corresponding text content and checks if the text contains hot words. If the hot words are present, the corresponding operation is executed.

[0052] To improve the accuracy of speech recognition, the selection of hot words to trigger corresponding operations is a crucial step. If a user speaks a voice command containing hot words from a preset hot word library and the system attempts to recognize it, it is easy to misidentify it as other words. Obviously, a system based on such a hot word library may fail to accurately recognize the user's voice commands during use, resulting in a poor user experience.

[0053] In related technologies, the impact of user pronunciation on hot word recognition is often overlooked when constructing hot word libraries. For example, similar pronunciations may correspond to different text content in different regional dialects, and the pronunciation of the same text content may also differ in different dialects. This can lead to situations where, after a user gives a voice command, the system does not respond in accordance with the user's intent, thus affecting the user experience. In such cases, building a hot word library solely based on the expected operation names may result in the inability to accurately identify hot words in the user's speech, thus failing to perform the correct interactive operation.

[0054] Furthermore, even within the same language environment, there are some words with similar pronunciations. The differences in pronunciation between these words are very subtle. Humans may be able to distinguish them through context and subtle tones, but this is a challenge for machines. When there are multiple words with similar pronunciations in the hot word library corresponding to the system, the user's voice data may be misidentified, which may then trigger the operation corresponding to other hot words incorrectly, causing confusion for the user. Therefore, the hot word library should try to avoid having a large number of hot words with similar pronunciations, or if it is unavoidable, consider replacing some words with synonyms.

[0055] In view of this, embodiments of this application provide a word addition method. Before adding the target word to a preset hot word library, the method considers the influence between the target word and words with similar pronunciations, as well as existing hot words in the preset hot word library. In this way, the probability of each hot word in the final hot word library being misidentified by each other is reduced to a certain extent, thereby improving the accuracy of recognizing speech data containing hot words and improving the user experience.

[0056] Please see Figure 1 , Figure 1 This is an application scenario diagram of the word addition method provided in the embodiments of this application. For example... Figure 1 As shown, the electronic device 10 can acquire the user's voice data and identify whether it includes hot words from a preset hot word library. If so, it executes the operation corresponding to the hot word. The electronic device 10 may include, but is not limited to, mobile phones, wearable devices (such as smartwatches, smart bracelets, smart glasses, etc.), tablets, laptops, in-vehicle terminals, PCs (Personal Computers), etc. The functions implemented by this method can be achieved by the processor in the electronic device calling program code. Of course, the program code can be stored in a computer storage medium. Therefore, the electronic device includes at least a processor and a storage medium.

[0057] The word addition method provided in this application includes: identifying multiple pronunciation-similar words of a target word in different linguistic environments, including a Mandarin environment and a target regional dialect environment; determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in the different linguistic environments, wherein the word to be compared includes the multiple pronunciation-similar words and multiple hot words contained in a preset hot word library; when the target word similarity meets a preset similarity condition, performing speech recognition processing on the speech data corresponding to the target word in different noise environments to obtain the speech recognition result of the target word; determining the comprehensive recognition reliability of the target word based on the target word similarity and the speech recognition result; and adding the target word to the preset hot word library when the comprehensive recognition reliability meets a preset reliability condition.

[0058] In some embodiments, the target word and the corresponding speech data of the word to be compared can be the user's speech data collected by the electronic device 10. The user's recorded speech data can help the user extract features based on their personal pronunciation characteristics such as voice and intonation, thereby helping the user add the target word to the preset hot word library to improve the accuracy of subsequent speech recognition.

[0059] In some embodiments, users can select either a Mandarin-speaking environment or a target regional dialect environment based on their own language context, and use the preset speech data of the words to be compared corresponding to the selected language environment to add the target words. This reduces the amount of user speech data collected and improves the efficiency of word addition. Furthermore, allowing users to select a language environment according to their needs and preferences makes speech recognition more aligned with their personal usage habits and language characteristics.

[0060] It should be noted that, as Figure 1 The application scenario diagrams of the word addition method shown are merely examples and not intended to limit the methods provided in this application. In some embodiments, the electronic device 10 may also be in the form of a server, which collects user voice data from associated devices distributed in different regions to obtain voice data corresponding to the target word and the voice data corresponding to the word to be compared in different language environments, and determines whether to add the target word to the preset hot word library using the word addition method provided in this application.

[0061] Furthermore, in some embodiments, after adding the target word to the preset hot word library, the word addition method provided in this application further includes: transmitting the constructed hot word library to other electronic devices that use the preset hot word library. In this way, multiple electronic devices can share the same hot word library, improving the efficiency of adding target words to the hot word library, and also facilitating the maintenance and modification of the hot word library, thus improving the applicability of the method.

[0062] The following is a flowchart illustrating the basic implementation of the word addition method provided in this application, with reference to the accompanying drawings.

[0063] Please see Figure 2 , Figure 2 This is a flowchart illustrating one implementation of the word addition method provided in an embodiment of this application. For example... Figure 2 As shown, the method may include the following steps 201 to 205:

[0064] Step 201: Identify multiple words with similar pronunciations to the target word in different linguistic contexts.

[0065] In this embodiment of the application, multiple words with similar pronunciations to the target word are identified in different linguistic environments, including Mandarin environment and target regional dialect environment.

[0066] It is understood that the different language environments in this application embodiment include Mandarin environment, or target regional dialect environments such as Min dialect environment, Hakka dialect environment, etc. Within different language environments or even within the same language environment, there exist words that are easily confused in pronunciation; in this application embodiment, these words are referred to as pronunciation-similar words. In the process of computer equipment recognizing speech data and obtaining text content, feature vectors containing sound characteristics are mainly obtained by extracting features from the speech data. The high feature similarity between the feature vectors of pronunciation-similar words may lead to confusion and misidentification by computer equipment.

[0067] To improve the accuracy of speech recognition for each hot word in the hot word library, the word addition method provided in this application first identifies multiple words with similar pronunciations in different linguistic environments for the target word to be added to the preset hot word library.

[0068] In some embodiments, the identified multiple words with similar pronunciations are words selected from different linguistic environments. In this way, when adding the target word to the preset hot word library, considering whether there are easily confused words with similar pronunciations in multiple different linguistic environments, the risk of misidentification of the target word can be more comprehensively assessed, the versatility of the constructed preset hot word library can be improved, and it can be adapted to various linguistic environments.

[0069] In some embodiments, for different language environments, multiple similar-sounding words corresponding to the target word in each different language environment are identified. Furthermore, when adding the target word to a preset hot word database, the decision to add the target word is made based on the specific language environment. This is because general users typically only perform speech recognition in one or a few language environments, such as a Mandarin-speaking environment and their local dialect. Determining whether to add the target word to the preset hot word database for each language environment improves the recognition accuracy and practicality in that specific language environment. It also allows users to choose their target language environment, thus improving recognition accuracy in a targeted manner.

[0070] To further improve user experience, in some embodiments, before determining multiple phonetically similar words of the target word in different linguistic environments, the method provided in this application further includes: determining the user's target linguistic environment. The target linguistic environment can be one or more linguistic environments, for example, selecting a Mandarin environment and a Hakka environment as the target linguistic environment. Thus, based on the selected target linguistic environment, multiple phonetically similar words of the target word are determined in the target environment. Furthermore, in subsequent steps, based on the phonetic similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in the target linguistic environment, the target word similarity between the target word and the word to be compared is determined.

[0071] In this way, based on user needs, the similarity of target words is confirmed only in a few language environments, which reduces the number of voice data of words to be compared, improves the efficiency of the word addition method, meets the user's voice recognition needs for hot words in the target language environment, and improves the response accuracy of the system based on the preset hot word library.

[0072] In some embodiments, determining multiple phonetically similar words of the target word in different linguistic environments includes:

[0073] In the different language environments, the spectral features of the speech data corresponding to the target word are compared with the spectral features of the speech data corresponding to words in the preset word library to determine words with similar pronunciations that have a higher feature similarity than a preset similarity threshold.

[0074] It should be noted that the preset vocabulary can be a set of words whose pronunciation is similar to the target word, selected based on human experience, or it can be a set of commonly used words obtained by processing historical speech data for speech recognition using machine learning and other algorithms. No limitation is made here.

[0075] The speech data corresponding to the target word and the words in the preset word library can be speech data containing the target word collected by a microphone. Multiple speech data samples from different users, different pronunciation methods or different background sound environments can be collected to improve the performance and robustness of this application in determining words with similar pronunciations.

[0076] In some embodiments, speech processing and feature extraction techniques are used on the speech data corresponding to the target word to convert the speech signal into a feature vector, thereby obtaining the spectral features corresponding to the target word. Commonly used feature extraction algorithms utilize spectral features such as Mel-frequency cepstral coefficients (MFCC) and linear-frequency cepstral coefficients (LFCC).

[0077] Mel-frequency cepstral coefficients (MFCCs) are cepstral coefficients based on the perceptual characteristics of the human ear, simulating the human auditory perception process. In speech signal processing, MFCCs are widely used in speech recognition, speech synthesis, and other fields. The extraction process of MFCCs is roughly as follows:

[0078] Pre-emphasis: The speech signal is pre-emphasized using a first-order differential filter to highlight the high-frequency components; Framing: The pre-emphasized speech signal is divided into several short frames, each typically 20-40 milliseconds long; Windowing: A window function (such as a Hamming window) is applied to each frame to increase the spectral resolution; Fast Fourier Transform (FFT): The windowed frame signal is subjected to a Fourier transform to obtain the spectrum of each frame; Mel filter bank: The spectrum is divided into multiple Mel bands to simulate the perceptual filtering effect of the human ear; Logarithmic energy: The logarithm of the spectral values ​​of each Mel band is taken to compress the dynamic range and improve numerical stability; Discrete cosine transform (DCT): The logarithmic spectrum of each Mel band is subjected to a discrete cosine transform to obtain the MFCC coefficients.

[0079] Through the above process, the speech signal is converted into a set of MFCC coefficients, which can effectively describe the characteristics of the speech signal and have good noise immunity.

[0080] Linear Frequency Cepstral Coefficients (LFCC) are another commonly used speech feature parameter. Their extraction process is similar to MFCC, but differs in some processing steps. During the framing stage, LFCC typically uses longer frame lengths, such as 40-60 milliseconds, to increase frequency resolution. Before the Mel filter bank, LFCC extracts linear predictive coefficients (LPC) by applying linear predictive coding (LPC) or complex cepstral analysis techniques. These coefficients are used to simulate the short-time predictability of the speech signal. LFCC uses LPC coefficients instead of the logarithmic spectrum in MFCC to perform a discrete cosine transform to obtain the LFCC coefficients. Compared to MFCC, LFCC has different characteristics in some aspects. For example, LFCC focuses more on simulating the dynamic characteristics of the speech signal, while MFCC focuses more on simulating the perceptual characteristics of the human ear.

[0081] In practical applications, appropriate feature parameters can be selected to process speech signals according to different needs and scenarios, and no restrictions are imposed here.

[0082] In some embodiments, the step of obtaining the spectral features of the speech data corresponding to the target word may include:

[0083] The speech data corresponding to the target word is divided into multiple frames according to a preset time interval;

[0084] Based on window functions, each frame of speech data is smoothed to reduce the impact of spectral leakage. Window functions include Hanning window, rectangular window, etc.

[0085] The smoothed speech data is subjected to a fast Fourier transform to convert the time domain signal into a frequency domain signal, thus obtaining the spectral energy distribution of each frame of speech data, i.e. the spectrum of each frame of speech data.

[0086] Each spectrum is filtered using a preset Mel filter to concentrate energy in the frequency range that the human ear is sensitive to, thus obtaining the Mel spectrum corresponding to each frame of speech data. The center frequency of the filter is set according to the Mel scale.

[0087] Based on the logarithmic function, the energy value of the filtered spectrum of the filter bank is logarithmically reduced and then compressed.

[0088] Non-static features are extracted from each continuous Mel spectrum to obtain the spectral features of the speech data corresponding to the target word. The preset algorithm includes first-order difference or second-order difference.

[0089] Through the above steps, the original speech data is converted into a quantifiable feature vector, and the spectral features of the speech data corresponding to the target word are obtained. The same method can be used to obtain the spectral features of the speech data corresponding to words in the preset word library. By comparing the feature similarity, words with feature similarity higher than the preset similarity threshold are identified as words with similar pronunciation.

[0090] Step 202: Determine the similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments.

[0091] In this embodiment of the application, the similarity between the target word and the word to be compared is determined based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments. The word to be compared includes the plurality of similar pronunciation words and the plurality of hot words contained in the preset hot word library.

[0092] It should be noted that the preset hot word library is a library that exists before the word addition method provided in this application is executed. When recognizing user voice data, the electronic device using the method provided in this application can determine the hot words in the preset hot word library after the word addition is completed and perform the corresponding response operation. The preset hot word library includes multiple hot words.

[0093] It is understandable that the similarity of the target word is determined based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared. The higher the pronunciation similarity between the target word and the word to be compared, the higher the similarity of the target word. At the same time, the risk of the target word being misidentified after being added to the preset hot word library is also higher. The method provided in this application should select target words with lower similarity to be added to improve the reliability of hot word addition.

[0094] Step 203: If the similarity of the target words meets the preset similarity conditions, perform speech recognition processing on the speech data corresponding to the target words in different noise environments to obtain the speech recognition results of the target words.

[0095] In this embodiment of the application, when the similarity of the target word meets the preset similarity condition, the speech data corresponding to the target word is subjected to speech recognition processing in different noise environments to obtain the speech recognition result of the target word.

[0096] It should be noted that different noise environments can include specified ambient sounds, such as human voice environments, music environments, or environments with specified decibel levels. By performing speech recognition processing in different noise environments, we can better simulate the actual application scenarios of users performing speech recognition, obtain speech recognition results for target words, and evaluate the impact of adding target words to a preset hot word library on the recognition of hot words, thereby improving the robustness of speech data containing hot words in recognition processing.

[0097] In some embodiments, the step of performing speech recognition processing on the speech data corresponding to the target word in different noise environments to obtain the speech recognition result of the target word includes:

[0098] Speech recognition processing is performed on the speech data corresponding to the target words in different noise environments to obtain the recognized text data of the speech data corresponding to the target words.

[0099] The recognized text data obtained in different noisy environments is matched with the target words to obtain speech recognition results including the number of successful matches and the success rate of matching.

[0100] By performing speech recognition processing on the speech data corresponding to target words in different noisy environments, a more comprehensive understanding of the speech recognition performance of target words in different environments can be obtained. It is understandable that, assuming the target word similarity obtained in the above steps remains constant and the total number of matches is fixed, the higher the number of successful matches and the higher the success rate of the match included in the speech recognition result, the higher the accuracy of speech recognition after the target word is added to the preset hot word library. Therefore, the decision to add the target word to the preset hot word library can be made based on the speech recognition result indicators.

[0101] In some embodiments, the speech recognition result also includes the number of recognition errors and the recognition results corresponding to different numbers of recognition errors.

[0102] By identifying the number of errors and the corresponding identification results for different numbers of errors, words with high similarity to the target word can be obtained. In some embodiments, after identifying words with high similarity to the target word and performing a comprehensive reliability assessment of the target word, if the target word is not added to the preset hot word library, it is replaced with semantically similar words. The word addition method provided in this application is then used to confirm whether the semantically similar words can be added to the preset hot word library. This method allows words that are easily misidentified but correspond to important operations to be added to the preset hot word library, ensuring the integrity of the preset hot word library.

[0103] For example, if the target word is "turn up the volume", and the method provided in this application determines that the overall recognition reliability of the target word does not meet the preset reliability conditions and cannot be added to the preset hot word library, the target word can be replaced with the semantically similar word "increase the volume", and it can be determined whether the semantically similar word can be added to the preset hot word library.

[0104] In some embodiments, when the similarity of the target words meets a preset similarity condition, performing speech recognition processing on the speech data corresponding to the target words in different noise environments to obtain the speech recognition result of the target words includes:

[0105] Speech recognition processing is performed on the speech data corresponding to the target words under different noise environments to obtain speech recognition results under each noise environment, including the number of recognition failures and the recognition failure rate.

[0106] Based on the speech recognition results under different noise environments, determine the number of noise environments in which the speech recognition results meet the preset recognition requirements. The noise environments that meet the preset recognition requirements have fewer recognition failures than a preset number threshold and a recognition failure rate less than a preset failure rate threshold.

[0107] If the number of environments exceeds a preset environment threshold, continue with the subsequent word addition steps.

[0108] Understandably, the more noisy environments there are where the number of recognition failures is less than a preset threshold and the recognition failure rate is less than a preset failure rate threshold, the more stable and reliable the target word's recognition performance will be in various noisy environments. Adding the target word to the preset hot word library can ensure that it can provide accurate recognition results in various different contexts, thereby improving the user experience.

[0109] Step 204: Determine the overall recognition reliability of the target words based on the similarity of the target words and the speech recognition results.

[0110] In this embodiment of the application, the overall recognition reliability of the target word is determined based on the similarity of the target word and the speech recognition result.

[0111] Understandably, the overall reliability of target word recognition is used to measure the accuracy of recognizing user speech data after adding target words to a preset hot word library. This includes whether the target words present in the speech data can be accurately identified, or whether they will be misidentified and cause erroneous responses when they are not present in the speech data. When the similarity between target words is low, and the accuracy of target word recognition is high in different noise environments represented by the speech recognition results, the overall reliability of target word recognition will also improve accordingly.

[0112] Step 205: If the overall recognition reliability meets the preset reliability conditions, add the target words to the preset hot word library.

[0113] In this embodiment of the application, if the overall recognition reliability meets the preset reliability conditions, the target word is added to the preset hot word library.

[0114] In some embodiments, the target word similarity is represented by a numerical value in the range of 0 to 1, and the indicators included in the speech recognition results, such as the number of successful matches and the success rate of matching, are also represented by numerical values. Multiple evaluation indicators are set based on the target word similarity and the speech recognition results. For example, if the target word similarity or the speech recognition result falls within a certain preset numerical range, a corresponding evaluation score is assigned to that indicator. Based on the preset weights of the target word similarity and the speech recognition results, an estimated comprehensive recognition reliability is obtained. When the comprehensive recognition reliability is higher than a preset reliability condition, the target word is added to a preset hot word library.

[0115] By implementing the above technical solution, the risk of misidentification of the target word is determined by the similarity between the target word and similar words in different language environments and existing hot words in the preset hot word library, as well as the result of speech recognition of the target word. This determines whether to add the target word to the preset hot word library, avoiding the situation where target words with high similarity to similar words in pronunciation and existing hot words in the preset word library are added to the preset hot word library, which could lead to errors in subsequent hot word recognition. This improves the accuracy of speech recognition of hot words in the preset word library after they have been added.

[0116] The following will explain, with reference to the accompanying drawings, the steps for determining the similarity between the target word and the word to be compared in the word addition method provided in this application.

[0117] Please see Figure 3 , Figure 3This is a flowchart illustrating another implementation of the word addition method provided in the embodiments of this application. For example... Figure 3 As shown, the method may include the following steps 301 to 307:

[0118] Step 301: Identify multiple words with similar pronunciations to the target word in different linguistic contexts.

[0119] Step 302: Based on the preset speech recognition algorithm, obtain the pronunciation similarity between the spectral features of the speech data corresponding to the target word in different language environments and the spectral features of the speech data corresponding to the word to be compared.

[0120] In some embodiments, determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments includes: obtaining the pronunciation similarity between the spectral features of the speech data corresponding to the target word and the spectral features of the speech data corresponding to the word to be compared in different language environments according to a preset speech recognition algorithm; obtaining the frequency of use of the word to be compared based on the frequency of occurrence of the word to be compared in a pre-stored historical text dataset, wherein the historical text dataset includes text data corresponding to multiple previously acquired speech data; and determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity and the frequency of use.

[0121] Based on a pre-defined speech recognition algorithm, the pronunciation similarity between the target word and multiple similarly pronounced words, as well as multiple hot words from a pre-defined hot word database, can be obtained in different language environments, based on their spectral features. Spectral features are important parameters reflecting speech characteristics in speech data; by comparing these features, the degree of similarity in pronunciation between two words can be assessed.

[0122] It should be noted that the preset speech recognition algorithm includes a spectral feature extraction algorithm and a similarity comparison algorithm. The spectral feature extraction algorithm can extract the spectral features of the speech data corresponding to the target word and the word to be compared. The methods for extracting common spectral features such as Mel-frequency cepstral coefficients (MFCC) and linear-frequency cepstral coefficients (LFCC) in step 201 above have been explained and will not be repeated here.

[0123] After obtaining the spectral features of the speech data corresponding to the target word and the speech data corresponding to the word to be compared, the similarity of these spectral features is compared using a similarity comparison algorithm. For example, the similarity of various dimensions of the spectral features, including frequency and amplitude, is compared, or the pronunciation similarity between the spectral features is obtained by calculating the Euclidean distance, cosine similarity, and other measures between the vectors of the spectral features.

[0124] It should be noted that, in some embodiments, multiple pronunciation similarities between the target word and multiple words to be compared in different speech environments are obtained, and the average of these pronunciation similarities is calculated to obtain the pronunciation similarity between the spectral features of the speech data corresponding to the target word and the spectral features of the speech data corresponding to the words to be compared, thus obtaining a more stable and reliable pronunciation similarity evaluation result.

[0125] Step 303: Based on the frequency of occurrence of the word to be compared in the pre-stored historical text dataset, obtain the frequency of use of the word to be compared.

[0126] In some embodiments, the frequency of use of the word to be compared is obtained based on the frequency of occurrence of the word in a pre-stored historical text dataset, wherein the historical text dataset includes text data corresponding to multiple previously acquired speech data.

[0127] It should be noted that the historical text dataset can be a dataset obtained based on user speech recognition, including text data corresponding to multiple previously acquired speech data. The historical text dataset can reflect how users use words in real-world application scenarios.

[0128] By statistically analyzing the frequency of occurrence of the words to be compared in historical text datasets, we can understand how the words are used in practical applications and thus assess their frequency of use. It can be understood that the higher the frequency of use, the higher the probability of the corresponding words to be compared appearing in subsequent speech recognition. If the pronunciation of the words to be compared is highly similar to that of the target words, it will affect the speech recognition results of the target words and increase the risk of misrecognition.

[0129] Step 304: Determine the target word similarity between the target word and the word to be compared based on pronunciation similarity and frequency of use.

[0130] In some embodiments, pronunciation similarity and frequency of use are estimated according to preset evaluation conditions, and then weighted according to different weights of pronunciation similarity and frequency of use to obtain the final target word similarity value. This weight can be determined according to the needs of the actual application and is not limited here.

[0131] Step 305: If the similarity of the target words meets the preset similarity conditions, perform speech recognition processing on the speech data corresponding to the target words in different noise environments to obtain the speech recognition results of the target words.

[0132] Step 306: Determine the overall recognition reliability of the target word based on the similarity of the target word and the speech recognition results.

[0133] Step 307: If the overall recognition reliability meets the preset reliability conditions, add the target words to the preset hot word library.

[0134] By implementing the above technical solution, this paper explains the method for determining the similarity of target words. It considers the phonetic similarity between words and utilizes historical datasets to analyze the frequency of use of the words to be compared. This method can more accurately determine the similarity of target words in practical applications, improving the accuracy and reliability of adding target words to a pre-set hot word database.

[0135] The following section, in conjunction with the accompanying drawings, explains another step in the word addition method provided in this application: determining the similarity between the target word and the word to be compared.

[0136] Please see Figure 4 , Figure 4 This is a flowchart illustrating another implementation of the word addition method provided in the embodiments of this application. For example... Figure 4 As shown, the method may include the following steps 401 to 406:

[0137] Step 401: Identify multiple words with similar pronunciations to the target word in different linguistic contexts.

[0138] Step 402: Determine the degree of misidentification of the target word based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to multiple words with similar pronunciations in different language environments.

[0139] In some embodiments, determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments includes: determining the degree of misidentification of the target word based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to a plurality of similarly pronounced words in different language environments, wherein the degree of misidentification is used to characterize the degree to which misidentification is easily caused due to the size of the pronunciation similarity; and, if the degree of misidentification meets a preset misidentification condition, determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to a plurality of hot words.

[0140] If the misidentification rate is high, it means that the target word should not be added to the preset hot word library. The process of determining the misidentification rate of the target word will be explained below.

[0141] In some embodiments, determining the degree of misidentification of the target word based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the plurality of similarly pronounced words in different language environments includes:

[0142] Based on the frequency of occurrence of the multiple similar-sounding words in a pre-stored historical text dataset, the frequency of use of the multiple similar-sounding words is obtained. The historical text dataset includes text data corresponding to multiple previously acquired speech data.

[0143] If the number of the plurality of similar-sounding words is less than a preset first number, at least some of the similar-sounding words among the plurality of similar-sounding words are determined to be popular similar words based on the frequency of use of different similar-sounding words;

[0144] If the number of popular similar words is less than a preset second quantity requirement, based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the multiple pronunciation similar words in different language environments, the number of misidentified language environments with a pronunciation similarity greater than a preset threshold is obtained.

[0145] Based on the number of misidentified environments, the pronunciation similarity between the speech data corresponding to the target word in different language environments and the speech data corresponding to the multiple similar-sounding words, and the frequency of use of different similar-sounding words, the overall misidentification situation corresponding to the multiple similar-sounding words is determined.

[0146] The degree of misidentification of the target word is determined based on the comprehensive misidentification situation, the number of words with similar pronunciations, and the number of popular similar words.

[0147] It should be noted that if the number of multiple words with similar pronunciations is greater than or equal to the preset first number, or the number of popular similar words is greater than or equal to the second number, then the target word has a large number of similar words or popular similar words, and the target word is at high risk of being misidentified, so it should not be added to the preset hot word database. Similarly, if the number of misidentified environments is too large, the target word should also not be added to the preset hot word database.

[0148] In some embodiments, weights can be set for different evaluation parameters through expert evaluation or machine learning to determine the overall misidentification of multiple words with similar pronunciations and the degree of misidentification of the target word.

[0149] Step 403: If the degree of misidentification meets the preset misidentification conditions, determine the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to multiple hot words.

[0150] In some embodiments, when the degree of misidentification meets the preset misidentification conditions, the similarity between the target word and the word to be compared is determined based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the plurality of hot words.

[0151] It should be noted that, in some embodiments, after determining the degree of misidentification based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to multiple words with similar pronunciations, the overall target word similarity between the target word and the word to be compared is determined by the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to multiple hot words. This provides a more comprehensive target word similarity assessment result.

[0152] Step 404: If the similarity of the target words meets the preset similarity conditions, perform speech recognition processing on the speech data corresponding to the target words in different noise environments to obtain the speech recognition results of the target words.

[0153] Step 405: Determine the overall recognition reliability of the target word based on the similarity of the target word and the speech recognition results.

[0154] Step 406: If the overall recognition reliability meets the preset reliability conditions, add the target words to the preset hot word library.

[0155] In some embodiments, after adding target words to a preset hot word library, the method provided in this application further includes training a classification model or a deep neural network model, wherein the classification model or deep neural network model is used to identify the differences between hot words and other words, thereby achieving accurate identification and triggering of hot words in user voice data, including:

[0156] Prepare training data: Use speech data samples labeled with different hot words from a pre-defined hot word library as training data. Ensure that each speech data sample is associated with a corresponding hot word for supervised learning;

[0157] Feature extraction: Feature extraction is performed on each speech data sample in the training data. Through a series of signal processing and feature extraction algorithms, the speech data samples are converted into quantifiable feature vectors. Commonly used feature extraction algorithms include Mel-frequency cepstral coefficients (MFCC) and linear predictive coding (LPC).

[0158] Data Labels: Add corresponding hot word labels to each feature vector and encode them to ensure that the model can learn the correspondence between the acoustic features of each hot word and the corresponding hot word labels;

[0159] Model selection and initialization: Select a model type suitable for the task, such as Support Vector Machine (SVM), Deep Neural Network (DNN), etc., and initialize the selected model;

[0160] Model training: The model is trained using labeled feature vectors and corresponding hot word tags.

[0161] In some embodiments, the trained model is optimized to improve the accuracy and robustness of hot word recognition, including: Dataset partitioning: dividing the training data into a training set and a validation set for training and evaluating the model; Model training and validation: training the model using the training set and evaluating its performance on the validation set, optimizing model parameters and structure; Evaluation metric selection: selecting appropriate evaluation metrics, such as accuracy and recall, based on task requirements; Performance evaluation: calculating the model's performance based on the evaluation metrics and identifying model problems; Model tuning: adjusting model parameters, such as learning rate and regularization coefficient, based on the problem analysis results to optimize model performance and generalization ability; Iterative repetition: retraining the model and evaluating its performance, continuously adjusting parameters and structure until optimal performance is achieved; Final model selection and saving: selecting the best-performing model as the final model and saving it for use in subsequent tasks.

[0162] In some embodiments, after model training and optimization are completed, users or system developers can add hot words to the speech recognition system as needed through a deployment operation. The deployment operation includes converting the model to a specific format and configuring system parameters to recognize specific hot words, including the following steps:

[0163] System testing and verification: Using voice data containing target hot words, the voice recognition system that integrates hot words is tested and verified to ensure that the system can accurately identify and trigger the operation or semantics corresponding to the target hot words;

[0164] Evaluate system performance: By testing and verifying the target hot words, evaluate the accuracy and performance of the system in identifying hot words, and determine whether further optimization and adjustment of the system is needed based on the evaluation results;

[0165] System Deployment: After completing testing, verification, and evaluation, the system with added hot words will be deployed to the actual environment to ensure that the system runs normally and can accurately identify and trigger hot word commands;

[0166] Continuous monitoring and maintenance: Regularly monitor and maintain the system to ensure the performance and stability of hot word recognition. Update and iterate the system as needed to meet user needs and improve system performance.

[0167] In summary, adding and deploying hot words in a speech recognition system is a key step in improving its functionality and user-friendliness. It helps to provide users with more personalized and customized services. For example, users can set personalized greetings or shortcuts for their electronic devices that have applied the word addition method provided in this application, thereby improving the convenience of interaction and enhancing the user experience.

[0168] It should be understood that although the steps in the above flowcharts 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 above flowcharts may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0169] Based on the foregoing embodiments, this application provides a word addition device, which includes various modules and units included in each module, and can be implemented by a processor; of course, it can also be implemented by specific logic circuits; in the implementation process, the processor can be a central processing unit (CPU), microprocessor (MPU), digital signal processor (DSP) or field programmable gate array (FPGA), etc.

[0170] Figure 5A schematic diagram of the word-adding device provided in the embodiments of this application is shown below. Figure 5 As shown, the word addition device 500 includes a similar word acquisition module 501, a similarity evaluation module 502, a reliability evaluation module 503, and an addition processing module 504, wherein:

[0171] The similar word acquisition module 501 is used to identify multiple pronunciation similar words of the target word in different linguistic environments, including the Mandarin environment and the dialect environment of the target region;

[0172] The similarity evaluation module 502 is used to determine the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments. The word to be compared includes the plurality of pronunciation similar words and the plurality of hot words contained in the preset hot word library.

[0173] The reliability assessment module 503 is used to perform speech recognition processing on the speech data corresponding to the target word in different noise environments when the similarity of the target word meets the preset similarity conditions, and obtain the speech recognition result of the target word; and determine the comprehensive recognition reliability of the target word based on the similarity of the target word and the speech recognition result.

[0174] An additional processing module 504 is used to add the target word to the preset hot word library when the overall recognition reliability meets the preset reliability conditions.

[0175] In some embodiments, the similar word acquisition module 501 is further configured to compare the spectral features of the speech data corresponding to the target word with the spectral features of the speech data corresponding to words in a preset word library under the different language environments, and determine words with a feature similarity higher than a preset similarity threshold as similar words in pronunciation.

[0176] In some embodiments, the similarity evaluation module 502 is further configured to: obtain the pronunciation similarity between the spectral features of the speech data corresponding to the target word in different language environments and the spectral features of the speech data corresponding to the word to be compared, according to a preset speech recognition algorithm; obtain the frequency of use of the word to be compared according to the frequency of occurrence of the word to be compared in a pre-stored historical text dataset, wherein the historical text dataset includes text data corresponding to multiple previously acquired speech data; and determine the target word similarity between the target word and the word to be compared based on the pronunciation similarity and the frequency of use.

[0177] In some embodiments, the similarity evaluation module 502 is further configured to determine the degree of misidentification of the target word based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the plurality of similar-sounding words in the different language environments, wherein the degree of misidentification is used to characterize the degree to which misidentification is easily caused due to the size of the pronunciation similarity; and when the degree of misidentification meets the preset misidentification conditions, determine the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the plurality of hot words.

[0178] In some embodiments, the similarity evaluation module 502 is further configured to: obtain the frequency of use of the plurality of similar-sounding words based on their frequency of occurrence in a pre-stored historical text dataset, wherein the historical text dataset includes text data corresponding to a plurality of previously acquired speech data; determine at least some of the similar-sounding words as popular similar words based on the frequency of use of different similar-sounding words when the number of the plurality of similar-sounding words is less than a preset first number; obtain the number of misidentified environments in which the speech data corresponding to the target word and the speech data corresponding to the plurality of similar-sounding words have a pronunciation similarity greater than a preset threshold based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the plurality of similar-sounding words in different language environments; determine the comprehensive misidentification situation of the plurality of similar-sounding words based on the number of misidentified environments, the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the plurality of similar-sounding words in different language environments, and the frequency of use of different similar-sounding words; and determine the degree of misidentification of the target word based on the comprehensive misidentification situation, the number of the plurality of similar-sounding words, and the number of popular similar words.

[0179] In some embodiments, the reliability assessment module 503 is further configured to perform speech recognition processing on the speech data corresponding to the target word in different noise environments to obtain the recognized text data of the speech data corresponding to the target word; and match the recognized text data obtained in different noise environments with the target word to obtain a speech recognition result including the number of successful matches and the proportion of successful matches.

[0180] The descriptions of the above device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0181] It should be noted that, in the embodiments of this application Figure 5The word addition device shown is illustrative of the module division, representing only one logical functional division; in actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, exist as separate physical entities, or be integrated into one unit with two or more units. The integrated units described above can be implemented in hardware, as software functional units, or a combination of software and hardware.

[0182] It should be noted that, in the embodiments of this application, if the above-described methods are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0183] This application provides a computer device, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing 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 an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the methods described above.

[0184] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method provided in the above embodiments.

[0185] This application provides a computer program product, including a computer program that, when executed by a processor, implements the method described in this application.

[0186] Those skilled in the art will understand that Figure 6The 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.

[0187] In one embodiment, the word addition device provided in this application can be implemented as a computer program, and the computer program can be implemented as follows: Figure 6 The device operates on the computer device shown. The memory of the computer device can store the various program modules that make up the above-described apparatus. The computer program, composed of the various program modules, causes the processor to execute the steps of the methods in the various embodiments of this application described in this specification.

[0188] It should be noted that the descriptions of the storage medium and device embodiments above are similar to the descriptions of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium, storage medium, and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0189] It should be understood that the phrases "one embodiment," "an embodiment," or "some embodiments" mentioned throughout the specification mean that a specific feature, structure, or characteristic related to an embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment," "in one embodiment," or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The descriptions of the various embodiments above tend to emphasize the differences between the various embodiments; their similarities or commonalities can be referred to mutually, and for the sake of brevity, they will not be repeated here.

[0190] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three kinds of relationships. For example, object A and / or object B can represent three situations: object A exists alone, object A and object B exist simultaneously, and object B exists alone.

[0191] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0192] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple modules or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or modules can be electrical, mechanical, or other forms.

[0193] The modules described above as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0194] In addition, each functional module in the various embodiments of this application can be integrated into one processing unit, or each module can be a separate unit, or two or more modules can be integrated into one unit; the integrated modules can be implemented in hardware or in the form of hardware plus software functional units.

[0195] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0196] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0197] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0198] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0199] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0200] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for adding words, characterized in that, Specifically, it includes: Identify multiple words with similar pronunciations to the target word in different linguistic environments, including Mandarin environments and dialect environments of the target region; Based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments, the target word similarity between the target word and the word to be compared is determined. The word to be compared includes the plurality of similar pronunciation words and the plurality of hot words contained in the preset hot word library. When the similarity of the target words meets the preset similarity conditions, the speech data corresponding to the target words is processed for speech recognition in different noise environments to obtain the speech recognition results of the target words. Based on the target word similarity and the speech recognition results, the overall recognition reliability of the target word is determined; If the overall recognition reliability meets the preset reliability conditions, the target word is added to the preset hot word library.

2. The method according to claim 1, characterized in that, The step of determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments includes: According to a preset speech recognition algorithm, the pronunciation similarity between the spectral features of the speech data corresponding to the target word in different language environments and the spectral features of the speech data corresponding to the word to be compared is obtained; Based on the frequency of occurrence of the word to be compared in the pre-stored historical text dataset, the frequency of use of the word to be compared is obtained. The historical text dataset includes text data corresponding to multiple previously acquired speech data. The similarity between the target word and the word to be compared is determined based on the pronunciation similarity and the frequency of use.

3. The method according to claim 1, characterized in that, The step of determining the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments includes: Based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the multiple similar-sounding words in different language environments, the degree of misidentification of the target word is determined. The degree of misidentification is used to characterize the degree to which misidentification is easily caused due to the size of the pronunciation similarity. If the degree of misidentification meets the preset misidentification conditions, the similarity between the target word and the word to be compared is determined based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the multiple hot words.

4. The method according to claim 3, characterized in that, The step of determining the degree of misidentification of the target word based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the plurality of similar-sounding words in different language environments includes: Based on the frequency of occurrence of the multiple similar-sounding words in a pre-stored historical text dataset, the frequency of use of the multiple similar-sounding words is obtained. The historical text dataset includes text data corresponding to multiple previously acquired speech data. If the number of the plurality of similar-sounding words is less than a preset first number, at least some of the similar-sounding words among the plurality of similar-sounding words are determined to be popular similar words based on the frequency of use of different similar-sounding words; If the number of popular similar words is less than a preset second quantity requirement, based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the multiple pronunciation similar words in different language environments, the number of misidentified language environments with a pronunciation similarity greater than a preset threshold is obtained. Based on the number of misidentified environments, the pronunciation similarity between the speech data corresponding to the target word in different language environments and the speech data corresponding to the multiple similar-sounding words, and the frequency of use of different similar-sounding words, the overall misidentification situation corresponding to the multiple similar-sounding words is determined. The degree of misidentification of the target word is determined based on the comprehensive misidentification situation, the number of words with similar pronunciations, and the number of popular similar words.

5. The method according to claim 1, characterized in that, The determination of multiple phonetically similar words for the target word in different linguistic contexts includes: In the different language environments, the spectral features of the speech data corresponding to the target word are compared with the spectral features of the speech data corresponding to words in the preset word library to determine words with similar pronunciations that have a higher feature similarity than a preset similarity threshold.

6. The method according to claim 1, characterized in that, The step of performing speech recognition processing on the speech data corresponding to the target word in different noise environments to obtain the speech recognition result of the target word includes: Speech recognition processing is performed on the speech data corresponding to the target words in different noise environments to obtain the recognized text data of the speech data corresponding to the target words. The recognized text data obtained in different noisy environments is matched with the target words to obtain speech recognition results including the number of successful matches and the success rate of matching.

7. A word addition device, characterized in that, include: The similar word acquisition module is used to identify multiple pronunciation similar words of the target word in different linguistic environments, including Mandarin environment and target regional dialect environment; The similarity assessment module is used to determine the target word similarity between the target word and the word to be compared based on the pronunciation similarity between the speech data corresponding to the target word and the speech data corresponding to the word to be compared in different language environments. The word to be compared includes the plurality of pronunciation similar words and the plurality of hot words contained in the preset hot word library. The reliability assessment module is used to perform speech recognition processing on the speech data corresponding to the target word in different noise environments when the similarity of the target word meets the preset similarity conditions, and obtain the speech recognition result of the target word; and determine the comprehensive recognition reliability of the target word based on the similarity of the target word and the speech recognition result. An additional processing module is added to add the target word to the preset hot word library when the overall recognition reliability meets the preset reliability conditions.

8. A computer device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images