Bluetooth headset based on laser engraving antenna and voice recognition method thereof
By using the laser engraving antenna technology in Bluetooth headsets for semantic feature recognition and phoneme analysis of speech signals, the problem of insufficient fuzzy speech recognition in the existing technology is solved, and efficient recognition and translation of foreign languages, dialects and ordinary speech is achieved, improving the efficiency of the recognition and translation of foreign languages, dialects and ordinary speech, The applicability and user experience of the system.
Patent Information
- Application Number
- CN202410985282.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-23
AI Technical Summary
The prior art fails to effectively identify fuzzy voice, reducing the usage rate of voice information and extending the information processing time.
Using Bluetooth headsets based on laser engraving antennas, the recognition type of speech signal is determined through semantic feature recognition and phoneme analysis, and the appropriate analysis method is selected according to the type to obtain the final translation results, including using language models or phoneme duplication recognition to improve recognition accuracy sex.
It improves the accuracy and speed of voice signals recognition, enhances the applicability and universality of the system, can process foreign languages, dialects and ordinary voice signals, provides more accurate and smooth translation services, and improves user experience and product performance.
Smart Images

Figure CN118540623B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of speech recognition, and in particular to a Bluetooth headset based on a laser-engraved antenna and a speech recognition method thereof. Background Art
[0002] With the continuous development of wireless communication technology and the popularization of smart devices, laser engraving earphones have received more and more attention and applications in the market. Laser engraving antenna technology can improve the efficiency of wireless signal reception and transmission of earphones, enhance signal stability and coverage, and enable users to get better audio experience and communication effect when using Bluetooth earphones.
[0003] Chinese Patent Publication No.: CN104103272A discloses a speech recognition method, device and Bluetooth headset, including: training command words in a command set, obtaining a reference feature vector sequence of each command word to construct a reference template for the command word, and the reference templates of each command word constitute a reference template set; extracting feature parameters from a voice signal to be tested, generating a feature vector sequence to be tested, and the feature vector sequence to be tested constitutes a test template; matching the test template with the reference template set to generate a matching score between the test template and each reference template; comparing the matching score of each reference template with a preset first threshold value and a preset second threshold value, if the difference between the second smallest score and the smallest score in the matching score of each reference template is greater than the second threshold value, and the smallest score is less than the first threshold value, then the command word corresponding to the reference template with the smallest score is recognized as the command word in the voice signal to be tested; it can be seen that the speech recognition method, device and Bluetooth headset have the following problems: no recognition processing is performed on ambiguous voice, which reduces the effective utilization rate of voice information and prolongs the information processing time. Summary of the invention
[0004] To this end, the present invention provides a Bluetooth headset based on a laser-engraved antenna and a voice recognition method thereof, so as to overcome the problem in the prior art that ambiguous voices are not recognized and processed, the effective utilization rate of voice information is reduced, and the information processing time is prolonged.
[0005] To achieve the above object, on the one hand, the present invention provides a method for voice recognition of a Bluetooth headset based on a laser-engraved antenna, comprising:
[0006] Step S1, the Bluetooth headset receives each voice signal sent by the sending device through the laser engraving antenna, and obtains the command signal sent by the sending device, the voice signal includes a foreign language signal, a dialect signal and a common voice signal;
[0007] Step S2, the Bluetooth headset recognizes semantic features according to each received voice signal segment, where the semantic features include language category, phonemes, and effectively recognized semantic words;
[0008] Step S3, determining the semantic recognition representation parameter of the current speech signal segment according to the number of effectively recognized semantic words and the total number of recognized phonemes in the current speech signal segment;
[0009] Step S4, determining the speech recognition type of the current speech signal segment according to the semantic recognition representation parameter of each of the speech signal segments;
[0010] Step S5, selecting an analysis method for the speech information contained in each speech signal segment according to the speech recognition type to determine the final translation result, including:
[0011] The language model is used to directly obtain the final translation result of the current speech signal segment according to all the effective recognition semantic words and the instruction signal.
[0012] Or, obtaining a number of preliminary semantic words in the speech signal segment according to the phonemes, obtaining a corresponding preliminary translation result containing each of the preliminary semantic words, calculating a semantic similarity representation parameter between each of the preliminary translation results and a final translation result corresponding to an adjacent speech signal segment that has completed the final translation, and selecting a final translation result for the speech signal segment according to the semantic similarity representation parameter;
[0013] Among them, the preparatory semantic word is the semantic word obtained after repeated recognition of a single non-effective semantic word;
[0014] Step S6: determining whether the current speech signal segment needs to be reacquired according to whether the final translation result exists and feedback information on the final translation result.
[0015] Furthermore, in step S2, the process of determining the effective recognition semantic words includes:
[0016] Determine all semantic words that are recognized,
[0017] If the language category of the recognized semantic word is the same as the language category of the speech signal segment, the recognized semantic word is determined as a valid recognized semantic word.
[0018] Furthermore, in step S3, the semantic recognition representation parameter is determined according to formula (1):
[0019]
[0020] In formula (1), K is the semantic recognition representation parameter, n is the total number of valid semantic words, and x i is the number of phonemes in the i-th valid recognition semantic word, X is the total number of phonemes in the current speech signal segment, and i is an integer greater than 0.
[0021] Furthermore, in the step S4, determining the speech recognition type of the current speech signal segment according to the semantic recognition representation parameter of each speech signal segment includes:
[0022] If the semantic recognition representation parameter is greater than the standard semantic recognition parameter, the speech recognition type of the current speech signal segment is determined to be a direct translation recognition type;
[0023] If the semantic recognition representation parameter is less than or equal to the standard semantic recognition parameter, the speech recognition type of the current speech signal segment is determined to be a fuzzy recognition type.
[0024] Furthermore, in the step S5, the analysis method for the speech information contained in each speech signal segment is selected according to the speech recognition type, including:
[0025] If the speech recognition type is a direct translation recognition type, then a language model is selected to directly obtain a final translation result of the current speech signal segment according to all the valid recognition semantic words and the instruction signal;
[0026] If the speech recognition type is a fuzzy recognition type, several preliminary semantic words in the speech signal segment need to be obtained based on phonemes, and the corresponding preliminary translation results containing each of the preliminary semantic words are obtained. The semantic similarity representation parameters of each of the preliminary translation results and the final translation result corresponding to the adjacent speech signal segment that completes the final translation are calculated, and the final translation result for the speech signal segment is selected according to the semantic similarity representation parameters.
[0027] Furthermore, the using of the language model to directly obtain the final translation result of the current speech signal segment according to all the effective recognition semantic words and the instruction signal includes:
[0028] Determining the language category of the final translation result according to the language category contained in the instruction signal;
[0029] The language model is used to translate the current speech signal into a final translation result according to the language category of the final translation result.
[0030] Furthermore, the semantic similarity representation parameter is determined according to formula (2):
[0031]
[0032] In formula (2), G is the semantic similarity representation parameter, is the sentence vector of the preliminary translation result, is the sentence vector of the final translation result of the adjacent speech signal segment, is the semantic relationship coefficient.
[0033] Furthermore, the selecting of the final translation result according to the semantic similarity representation parameter includes:
[0034] If there is a preliminary semantic word whose semantic similarity representation parameter satisfies the translation condition, it is determined that there is a final translation result, and the final translation result is determined according to the preliminary semantic word with the largest absolute value of the semantic similarity representation parameter;
[0035] If there is no preliminary semantic word whose semantically similar representation parameter satisfies the translation condition, it is determined that there is no final translation result;
[0036] The translation condition is that the absolute value of the semantic similarity representation parameter is greater than the standard similarity representation parameter.
[0037] Furthermore, in step S6, if it is determined that there is no final translation result or the user feedback indicates that the final translation result is wrong, it is determined that the current speech signal segment needs to be re-acquired.
[0038] On the other hand, the present invention also provides a Bluetooth headset based on a laser-engraved antenna, comprising: a laser-engraved signal receiving module, which includes a laser-engraved antenna circuit board for receiving a voice signal and a command signal sent by a sending device;
[0039] A voice feature recognition module, which is connected to the laser engraving signal receiving module and is used to perform semantic feature recognition on each voice signal segment;
[0040] A speech classification module, which is connected to the speech feature recognition module, and is used to determine a semantic recognition representation parameter according to the recognition result of the semantic feature, and to determine the speech recognition type of each speech signal segment according to the semantic recognition representation parameter;
[0041] A speech analysis module, which is connected to the speech classification module and the speech feature module respectively, and is used to select an analysis method for the speech information contained in each speech signal segment according to the speech recognition type to determine the final translation result of the speech signal;
[0042] The basic function module is connected to the voice analysis module and the laser engraving signal receiving module respectively, and is provided with a processor and a broadcasting device to realize the basic functions of the Bluetooth headset and play the voice translation result.
[0043] Compared with the prior art, the beneficial effects of the present invention are that the present invention can process foreign languages, dialects and ordinary voice signals, improving the applicability and universality of the system; through semantic features and phoneme analysis, the recognition accuracy and speed of voice signals are improved; according to user feedback and usage, the system can effectively select the final translation result through a variety of analysis methods, improving the accuracy and fluency of the translation. In general, the system combines a number of technical advantages to provide users with more convenient and accurate voice recognition and translation services; laser engraving antenna technology brings significant advantages to portable devices such as Bluetooth headsets. By directly forming a three-dimensional circuit on the product shell, it not only saves space and improves design freedom, but also enhances the efficiency of signal reception and transmission, reduces interference, and improves the aesthetics of the product. At the same time, it supports accurate signal processing and multiple voice recognition in complex environments, thereby significantly improving user experience and product performance.
[0044] Furthermore, the received language model is divided into two speech recognition types, and corresponding analysis methods are provided, and different language types are processed using different language methods, thereby improving the accuracy of the translation results while ensuring the processing rate; for the direct translation recognition type, it is suitable for receiving clear and accurate speech signals, and does not require complex processing or analysis, because the speech signal has high quality and is easily directly recognized and translated by the language model at a fast speed. The translation result is basically a direct interpretation of the speech signal, saving processor computing power; for the fuzzy recognition type, it is necessary to re-acquire several preparatory semantic words by repeatedly recognizing the phonemes of non-validly recognized semantic words, and use the language model for further analysis and processing to obtain the final translation result. This analysis method can provide more accurate translation results in complex environments.
[0045] Furthermore, in the present invention, a sentence vector model is used to determine the semantic relationship between adjacent speech signals. By calculating the similarity between two sentence vectors, the semantic relationship between them can be quantified, including synonymy, correlation, etc. The sentence vector model can help understand the meaning of sentences in different contexts, and combine the word order, grammatical structure and contextual information in the sentence to more comprehensively capture the meaning of the sentence. It can also perform text similarity calculations to evaluate the semantic proximity of two sentences, thereby obtaining more accurate translation results.
[0046] Furthermore, if there are multiple preliminary semantic words at the same time, the final selected preliminary semantic word is determined according to the absolute values of the similarity representation parameters of each preliminary semantic word. The larger the absolute value of the similarity representation parameter, the more the semantic feature corresponding to the current preliminary semantic word is in line with the language logic of the semantics of the previous speech signal. Therefore, the preliminary semantic word with the largest absolute value of the similarity representation parameter is selected as the final translation result to ensure the accuracy and rationality of the Bluetooth headset translation. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 This is a flow chart of a method for voice recognition of a Bluetooth headset based on a laser-engraved antenna according to an embodiment of the present invention;
[0048] Figure 2 A logic diagram for determining speech recognition type for an embodiment of the present invention;
[0049] Figure 3 A logic diagram for determining whether a final translation result exists in an embodiment of the present invention;
[0050] Figure 4 The figure is a schematic diagram of the structure of a Bluetooth headset based on a laser-engraved antenna according to an embodiment of the present invention. DETAILED DESCRIPTION
[0051] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0052] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.
[0053] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.
[0054] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0055] Laser-cut antenna (LDS antenna) is a method of directly forming a metal antenna on the product shell using laser direct forming technology. Compared with traditional antennas, it has significant beneficial effects, including: high space utilization: by integrating the antenna into the product shell, the internal space occupied can be reduced, making the product design more compact and improving space utilization; strong performance stability: since the antenna is formed directly on the shell, the interference of internal components is reduced, thereby increasing the stability of the antenna; strong adaptability: LDS antenna technology can be applied to products of various shapes and structures, providing design flexibility; high signal quality: due to the direct integration of the antenna with the product shell, it helps to improve the quality of signal reception and transmission; reduce external interference: placing the antenna in the product appearance area can reduce the interference of the internal circuit on the antenna signal, thereby improving the purity of the signal.
[0056] See also Figure 1 As shown, it is a flow chart of the voice recognition method of the Bluetooth headset based on the laser engraving antenna in this embodiment of the present invention. To achieve the above-mentioned purpose, on the one hand, the present invention provides a voice recognition method of the Bluetooth headset based on the laser engraving antenna, comprising:
[0057] Step S1, the Bluetooth headset receives each voice signal sent by the sending device through the laser engraving antenna, and obtains the command signal sent by the sending device, the voice signal includes a foreign language signal, a dialect signal and a common voice signal;
[0058] Step S2, the Bluetooth headset recognizes semantic features according to each received voice signal segment, where the semantic features include language category, phonemes, and effectively recognized semantic words;
[0059] Step S3, determining a semantic recognition representation parameter of the current voice signal segment according to the number of effectively recognized semantic words in the current voice signal segment and the total number of phonemes recognized by the laser engraving antenna;
[0060] Step S4, determining the speech recognition type of the current speech signal segment according to the semantic recognition representation parameter of each of the speech signal segments;
[0061] Step S5, selecting an analysis method for the speech information contained in each speech signal segment according to the speech recognition type to determine the final translation result, including:
[0062] The language model is used to directly obtain the final translation result of the current speech signal segment according to all the effective recognition semantic words and the instruction signal.
[0063] Or, obtaining a number of preliminary semantic words in the speech signal segment according to the phonemes, obtaining a corresponding preliminary translation result containing each of the preliminary semantic words, calculating a semantic similarity representation parameter between each of the preliminary translation results and a final translation result corresponding to an adjacent speech signal segment that has completed the final translation, and selecting a final translation result for the speech signal segment according to the semantic similarity representation parameter;
[0064] Among them, the preparatory semantic word is the semantic word obtained after repeated recognition of a single non-effective semantic word;
[0065] Step S6: determining whether to reacquire the current speech signal segment according to whether the final translation result exists and feedback information on the final translation result.
[0066] It can be understood that the adjacent speech signal segment is the previous speech signal segment of the current speech signal segment.
[0067] In this embodiment, the semantic feature recognition of each speech signal segment by the processor can be obtained by any language model and acoustic model. For example, useful features such as Mel-frequency cepstral coefficients (MFCCs), pitch, sound intensity, etc. are extracted from the preprocessed speech signal, and the extracted features are matched with possible acoustic units (such as phonemes or letters) using an acoustic model (such as a hidden Markov model HMM or a deep learning model); and the probability information of the occurrence of words, phrases or sentences provided in the language model is used to determine the most likely semantic content; in a speech recognition system, a language model is usually used in combination with an acoustic model to improve the accuracy of recognition; this is a prior art and will not be described in detail;
[0068] The command signal includes whether to translate the speech signal and the language category obtained by translating the speech signal; the semantic features also include the characteristic parameters commonly used in automatic speech recognition (ASR) systems such as pitch, duration, and intensity;
[0069] The feedback information for the final translation result is the voice information sent by the user regarding whether the final translation result meets the requirements, which can be directly recognized by the language model and will not be elaborated here.
[0070] In this embodiment, the sending device is a device that can be connected to a Bluetooth headset and send out signals, such as a mobile phone, a computer, etc.
[0071] Specifically, in step S2, the process of determining the effective recognition semantic words includes:
[0072] Determine all the semantic words that are recognized,
[0073] If the language category of the recognized semantic word is the same as the language category of the speech signal segment, the recognized semantic word is determined as a valid recognized semantic word.
[0074] In implementation, semantic words are determined by identifying phonemes in combination with acoustic and language models;
[0075] The language model and acoustic model used in this embodiment are any of the existing models and are not specifically limited here.
[0076] Phonemes are the basic units of speech, and semantic words are the smallest units that carry meaning in a language; different language categories contain different numbers of phonemes.
[0077] In the present invention, an acoustic model is used to match phonemes with features of speech signals. The acoustic model is usually based on statistics or deep learning technology and is used to predict the sound features corresponding to a series of phonemes to obtain semantic words that match the phonemes; a language model is then used to predict the probability of occurrence of a word sequence based on statistical laws in the language, and to identify words that are more likely in a specific context; and a decoding algorithm is combined with the outputs of the acoustic model and the language model, and a technique such as the Viterbi algorithm is used to convert the phoneme sequence into the most likely word sequence; the decoded word sequence is matched with a built-in dictionary in the model to determine the valid semantic words that are finally recognized.
[0078] See also Figure 2 As shown, it is a logic diagram for determining the speech recognition type according to an embodiment of the present invention. In step S3, the semantic recognition representation parameter is determined according to formula (1):
[0079]
[0080] In formula (1), K is the semantic recognition representation parameter, n is the total number of valid semantic words, and x i is the number of phonemes in the i-th valid recognition semantic word, X is the total number of phonemes in the current speech signal segment, and i is an integer greater than 0.
[0081] In the present invention, considering that the sound received by the earphone may include the user's voice mixed with the noise in the environment, the voices of other people near the user, and the user's inability to utter clear language information due to his / her own reasons, which may interfere with the semantic recognition of the earphone, it is more reasonable and accurate to use phonemes to judge the accuracy of recognition, which can characterize the interference of abnormal sounds emitted by the user or other sounds in the environment on the semantic recognition of the earphone, and can better adaptively select subsequent analysis methods.
[0082] Specifically, in the step S4, determining the speech recognition type of the current speech signal segment according to the semantic recognition representation parameter of each speech signal segment includes:
[0083] If the semantic recognition representation parameter is greater than the standard semantic recognition parameter, the speech recognition type of the current speech signal segment is determined to be a direct translation recognition type;
[0084] If the semantic recognition representation parameter is less than or equal to the standard semantic recognition parameter, the speech recognition type of the current speech signal segment is determined to be a fuzzy recognition type.
[0085] In the implementation, the standard semantic recognition parameters are selected in the interval [0.8, 0.95];
[0086] It is understandable that the lengths of speech signal segments are different, and the total number of phonemes in different speech signal segments will vary greatly. The total number of phonemes in long speech signal segments is greater than that in short speech signal segments. Therefore, during the recognition process, the number of interfering phonemes in the language contained in long speech signal segments is greater than that in short speech signal segments. In actual applications, the standard semantic recognition parameters of short speech signal segments are greater than those of long speech signal segments.
[0087] Specifically, in step S5, the analysis method for the speech information contained in each speech signal segment is selected according to the speech recognition type, including:
[0088] If the speech recognition type is a direct translation recognition type, then a language model is selected to directly obtain a final translation result of the current speech signal segment according to all the valid recognition semantic words and the instruction signal;
[0089] If the speech recognition type is a fuzzy recognition type, several preliminary semantic words in the speech signal segment need to be obtained based on phonemes, and the corresponding preliminary translation results containing each of the preliminary semantic words are obtained. The semantic similarity representation parameters of each of the preliminary translation results and the final translation result corresponding to the adjacent speech signal segment that completes the final translation are calculated, and the final translation result for the speech signal segment is selected according to the semantic similarity representation parameters.
[0090] It is understandable that the voice signal received by the Bluetooth headset in the direct translation recognition type is clearly identifiable and is language information with little interference from the external environment, so it does not require excessive processing. However, the fuzzy recognition type cannot be directly and clearly recognized due to factors such as interference that cannot be eliminated from the external environment or the user's own unclear pronunciation. The received voice signal requires certain processing and analysis to obtain the final translation result.
[0091] In the present invention, the received language model is divided into two speech recognition types, and corresponding analysis methods are provided. Different language types are processed using different language methods, which improves the accuracy of the translation result while ensuring the processing rate; for the direct translation recognition type, it is suitable for receiving clear and accurate speech signals, and does not require complex processing or analysis. Because the quality of the speech signal is high, it is easy to be directly recognized and translated by the language model at a fast speed. The translation result is basically a direct interpretation of the speech signal, saving the computing power of the processor; for the fuzzy recognition type, it is necessary to re-acquire several preparatory semantic words by repeatedly recognizing the phonemes of non-valid recognized semantic words, and use the language model for further analysis and processing to obtain the final translation result. This analysis method can provide more accurate translation results in complex environments.
[0092] Specifically, the method of using the language model to directly obtain the final translation result of the current speech signal segment according to all the effective recognition semantic words and the instruction signal includes:
[0093] Determining the language category of the final translation result according to the language category contained in the instruction signal;
[0094] The language model is used to translate the current speech signal into a final translation result according to the language category of the final translation result.
[0095] See also Figure 3 As shown, it is a logic diagram for determining whether there is a final translation result in an embodiment of the present invention. The semantic similarity representation parameter is determined according to formula (2):
[0096]
[0097] In formula (2), G is the semantic similarity representation parameter, is the sentence vector of the preliminary translation result, is the sentence vector of the final translation result of the adjacent speech signal segment, is the semantic relationship coefficient.
[0098] Sentence Vector Model is a natural language processing (NLP) technology used to map sentences or short texts into a continuous vector space. The goal of these models is to capture the semantic information of sentences and enable them to be compared and calculated in the vector space to facilitate various NLP tasks such as text similarity calculation, classification, clustering, etc.
[0099] The sentence vector model can capture the semantic content of a sentence, making semantically similar sentences closer in the vector space. The sentences are converted into vectors of fixed length to facilitate processing by machine learning algorithms. The vectorized representation makes it possible to quickly calculate the similarity between sentences or perform other mathematical operations. The sentence vector model can be used for a variety of NLP tasks, including but not limited to text classification, sentiment analysis, question-answering systems, etc.
[0100] Cosine similarity is a measure of the similarity of the angle between vectors, with a value range of -1 to 1. Represented as sentence vector With sentence vector The cosine similarity of the two sentences is greater, the closer the relationship between the two sentences is.
[0101] σ is the semantic relationship coefficient, which takes values of -1 or 1. The value of the semantic relationship coefficient is determined by whether there are transitional relationship words in the effective recognition semantic words, such as but, yet, and yet. When transitional relationship words are included, σ takes the value of -1, and when transitional relationship words are not included, σ takes the value of 1.
[0102] In the present invention, a sentence vector model is used to determine the semantic relationship between adjacent speech signals. By calculating the similarity between two sentence vectors, the semantic relationship between them can be quantified, including synonymy, correlation, etc. The sentence vector model can help understand the meaning of sentences in different contexts, and combine the word order, grammatical structure and contextual information in the sentence to more comprehensively capture the meaning of the sentence. It can also perform text similarity calculation to evaluate the semantic proximity of two sentences, thereby obtaining a more accurate translation result.
[0103] Specifically, the selection of the final translation result according to the semantic similarity representation parameter includes:
[0104] If there is a preliminary semantic word whose semantic similarity representation parameter satisfies the translation condition, it is determined that there is a final translation result, and the final translation result is determined according to the preliminary semantic word with the largest absolute value of the semantic similarity representation parameter;
[0105] If there is no preliminary semantic word whose semantically similar representation parameter satisfies the translation condition, it is determined that there is no final translation result;
[0106] The translation condition is that the absolute value of the semantic similarity representation parameter is greater than the standard similarity representation parameter.
[0107] In the implementation, the standard similarity characterization parameters are selected in the interval [0.5, 0.8];
[0108] The preliminary semantic words are semantic words obtained after repeated recognition of a single non-valid semantic word. Generally, the number is not unique. By performing secondary fuzzy recognition on the non-valid semantic words, semantic words with the same language category as the speech signal segment are obtained, which are the preliminary semantic words. Fuzzy recognition is the recognition of phonemes in combination with the current language category and the acoustic model and language model. This is a prior art and will not be elaborated on.
[0109] In this implementation, if there are multiple preparatory semantic words at the same time, the final selected preparatory semantic word is determined based on the absolute values of the similarity representation parameters of each preparatory semantic word. The larger the absolute value of the similarity representation parameter, the more the semantic feature corresponding to the current preparatory semantic word is in line with the language logic of the semantics of the previous speech signal. Therefore, the preparatory semantic word with the largest absolute value of the similarity representation parameter is selected as the final translation result to ensure the accuracy and rationality of the Bluetooth headset translation.
[0110] Specifically, in step S6, if it is determined that there is no final translation result or the user feedback indicates that the final translation result is wrong, it is determined that the current speech signal segment needs to be re-acquired.
[0111] See also Figure 4 As shown, it is a schematic diagram of the structure of a Bluetooth headset based on a laser engraving antenna according to an embodiment of the present invention. The present invention also provides a Bluetooth headset based on a laser engraving antenna, including:
[0112] A laser engraving signal receiving module, which includes a laser engraving antenna circuit board, for receiving voice signals and command signals sent by a sending device;
[0113] A voice feature recognition module, which is connected to the laser engraving signal receiving module and is used to perform semantic feature recognition on each voice signal segment;
[0114] A speech classification module, which is connected to the speech feature recognition module, and is used to determine a semantic recognition representation parameter according to the recognition result of the semantic feature, and to determine the speech recognition type of each speech signal segment according to the semantic recognition representation parameter;
[0115] A speech analysis module, which is connected to the speech classification module and the speech feature module respectively, and is used to select an analysis method for the speech information contained in each speech signal segment according to the speech recognition type to determine the final translation result of the speech signal;
[0116] The basic function module is connected to the voice analysis module and the laser engraving signal receiving module respectively, and is provided with a processor and a broadcasting device to realize the basic functions of the Bluetooth headset and play the voice translation result.
[0117] Laser Direct Structuring (LDS) antennas are a technology that uses laser technology to directly form three-dimensional circuits on the surface of materials such as plastics. LDS technology allows the antenna structure to be directly integrated into the product's housing or internal structure during the product design and manufacturing process, thereby saving space and increasing design flexibility.
[0118] Bluetooth headsets based on laser antennas use laser technology to directly form metal antennas on the headset housing. This design provides greater design flexibility and improves RF performance, making Bluetooth connections farther and more stable. LDS technology uses lasers to irradiate specific materials, releasing metal particles to form the required antenna structure. The advantage of this method is that it can directly form fine metal patterns on three-dimensional plastic parts without the need for additional antenna components, which helps reduce the size and weight of the headset while improving signal transmission efficiency and stability.
[0119] In this embodiment, the laser engraving signal receiving module includes a laser engraving antenna circuit board, and the laser engraving antenna circuit board can be set at any position inside the earphone, preferably on the inner surface of the earphone shell to save the internal space of the earphone shell;
[0120] The present invention does not limit the specific structure of the speech feature recognition module, the speech classification module and the speech analysis module, which can be composed of logic components, and the logic components can be any one of the existing microprocessors containing language models, which are not specifically limited here;
[0121] The basic function module includes earphones that can realize basic broadcasting functions. This is a prior art and is not specifically limited here.
[0122] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.
[0123] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A speech recognition method for Bluetooth headset based on laser engraving antenna, It is characterized in that include: Step S1, the Bluetooth headset receives each voice signal sent by the sending device through the laser engraving antenna, and obtains the command signal sent by the sending device, the voice signal includes a foreign language signal, a dialect signal and a common voice signal; Step S2, the Bluetooth headset recognizes semantic features according to each received voice signal segment, where the semantic features include language category, phonemes, and effectively recognized semantic words; Step S3, determining the semantic recognition representation parameter of the current speech signal segment according to the number of effectively recognized semantic words and the total number of recognized phonemes in the current speech signal segment; Step S4, determining the speech recognition type of the current speech signal segment according to the semantic recognition representation parameter of each of the speech signal segments; Step S5, selecting an analysis method for the speech information contained in each speech signal segment according to the speech recognition type to determine the final translation result, including: The language model is used to directly obtain the final translation result of the current speech signal segment according to all the effective recognition semantic words and the instruction signal. Or, obtaining a number of preliminary semantic words in the speech signal segment according to the phonemes, obtaining a corresponding preliminary translation result containing each of the preliminary semantic words, calculating a semantic similarity representation parameter between each of the preliminary translation results and a final translation result corresponding to an adjacent speech signal segment that has completed the final translation, and selecting a final translation result for the speech signal segment according to the semantic similarity representation parameter; Among them, the preparatory semantic word is the semantic word obtained after repeated recognition of a single non-effective semantic word; Step S6: determining whether the current speech signal segment needs to be reacquired according to whether the final translation result exists and feedback information on the final translation result.
2. The speech recognition method of the Bluetooth headset based on the laser engraving antenna according to claim 1, It is characterized in that In step S2, the process of determining the effective recognition semantic words includes: Determine all semantic words that are recognized, If the language category of the recognized semantic word is the same as the language category of the speech signal segment, the recognized semantic word is determined as a valid recognized semantic word.
3. The speech recognition method of the Bluetooth headset based on the laser engraving antenna according to claim 2, It is characterized in that In step S3, the semantic recognition representation parameter is determined according to formula (1): , In formula (1), K is the semantic recognition representation parameter, n is the total number of valid semantic words, and x i is the number of phonemes in the i-th valid recognition semantic word, X is the total number of phonemes in the current speech signal segment, and i is an integer greater than 0.
4. The voice recognition method of the Bluetooth headset based on the laser engraving antenna according to claim 3, It is characterized in that In the step S4, determining the speech recognition type of the current speech signal segment according to the semantic recognition representation parameter of each speech signal segment includes: If the semantic recognition representation parameter is greater than the standard semantic recognition parameter, the speech recognition type of the current speech signal segment is determined to be a direct translation recognition type; If the semantic recognition representation parameter is less than or equal to the standard semantic recognition parameter, the speech recognition type of the current speech signal segment is determined to be a fuzzy recognition type.
5. The method for voice recognition of a Bluetooth headset based on a laser-engraved antenna according to claim 4, It is characterized in that In the step S5, the analysis method for the speech information contained in each speech signal segment is selected according to the speech recognition type, including: If the speech recognition type is a direct translation recognition type, then a language model is selected to directly obtain a final translation result of the current speech signal segment according to all the valid recognition semantic words and the instruction signal; If the speech recognition type is a fuzzy recognition type, several preliminary semantic words in the speech signal segment need to be obtained based on phonemes, and the corresponding preliminary translation results containing each of the preliminary semantic words are obtained. The semantic similarity representation parameters of each of the preliminary translation results and the final translation result corresponding to the adjacent speech signal segment that completes the final translation are calculated, and the final translation result for the speech signal segment is selected according to the semantic similarity representation parameters.
6. The method for voice recognition of a Bluetooth headset based on a laser-engraved antenna according to claim 5, It is characterized in that The method of using the language model to directly obtain the final translation result of the current speech signal segment according to all the effective recognition semantic words and the instruction signal includes: Determining the language category of the final translation result according to the language category contained in the instruction signal; The language model is used to translate the current speech signal into a final translation result according to the language category of the final translation result.
7. The method for voice recognition of a Bluetooth headset based on a laser-engraved antenna according to claim 6, It is characterized in that The semantic similarity representation parameter is determined according to formula (2): , In formula (2), G is the semantic similarity representation parameter, is the sentence vector of the preliminary translation result, is the sentence vector of the final translation result of the adjacent speech signal segment, is the semantic relationship coefficient.
8. The method for voice recognition of a Bluetooth headset based on a laser-engraved antenna according to claim 7, It is characterized in that The selecting of the final translation result according to the semantic similarity representation parameter includes: If the preset translation conditions are met, it is determined that there is a final translation result, and the preliminary translation result corresponding to the maximum semantic similarity representation parameter is determined as the final translation result; If there is no preliminary semantic word whose semantically similar representation parameter satisfies the translation condition, it is determined that there is no final translation result; The translation condition is that the absolute value of the semantic similarity representation parameter is greater than the standard similarity representation parameter.
9. The method for voice recognition of a Bluetooth headset based on a laser-engraved antenna according to claim 8, It is characterized in that In step S6, if it is determined that there is no final translation result or the user feedback indicates that the final translation result is wrong, it is determined that the current speech signal segment needs to be re-acquired.
10. A Bluetooth headset based on a laser-engraved antenna using the speech recognition method according to any one of claims 1 to 9, It is characterized in that include: A laser engraving signal receiving module, which includes a laser engraving antenna circuit board, for receiving voice signals and command signals sent by a sending device; A voice feature recognition module, which is connected to the laser engraving signal receiving module and is used to perform semantic feature recognition on each voice signal segment; A speech classification module, which is connected to the speech feature recognition module, and is used to determine a semantic recognition representation parameter according to the recognition result of the semantic feature, and to determine the speech recognition type of each speech signal segment according to the semantic recognition representation parameter; A speech analysis module, which is connected to the speech classification module and the speech feature recognition module respectively, and is used to select an analysis method for the speech information contained in each speech signal segment according to the speech recognition type to determine the final translation result of the speech signal; The basic function module is connected to the voice analysis module and the laser engraving signal receiving module respectively, and is provided with a processor and a broadcasting device to realize the basic functions of the Bluetooth headset and play the voice translation result.
