Self-adaptive multilingual inter-translation system

Abstract

The invention discloses a self-adaptive multi-language inter-translation system, and relates to the technical field of multi-language inter-translation, and the self-adaptive multi-language inter-translation system comprises a voice acquisition module, a language analysis module, a semantic recognition module, a bidirectional translation module and a self-adaptive learning module. The translation process is continuously optimized by analyzing the user feedback and the translation result, so that the translation quality can be continuously improved along with time and data accumulation, meanwhile, the forward and reverse translation function provided by the bidirectional translation module ensures that the user can be well supported whether inputting or receiving information, and the user experience is improved. A user can switch languages at any time according to needs, so that communication is smoother, and free flow of information and communication of multi-culture are promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-language translation, and in particular to an adaptive multi-language translation system. Background Art

[0002] As globalization continues to deepen, language diversity often becomes a barrier to communication. In order to break this barrier, it is particularly important to establish an adaptive multilingual translation system. This system not only supports the input and output of multiple languages, but also ensures the accuracy of conversion between various languages. By adopting advanced natural language processing technology and machine learning algorithms, this platform can learn and adapt to new language usage and expressions in real time. Through such a multilingual translation system, we can more effectively connect users around the world, promote the free flow of culture and information, and continuously improve the quality of translation.

[0003] The existing technology is not convenient for supporting translation between multiple languages, is not convenient for serving a wider user group, and is not convenient for optimizing the translation process according to user feedback and translation results, thereby reducing the accuracy of translation and the user experience. At the same time, it is not convenient to perform key feature analysis on the input target audio signal, which seriously affects the accuracy of the translation and leads to translation errors, such as incorrect interpretation of the meaning or context of a word, which in turn makes the translation content inaccurate or easy to cause misunderstanding.

[0004] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0005] In view of the problems in the related art, the present invention proposes an adaptive multi-language translation system to overcome the above technical problems existing in the existing related art.

[0006] To this end, the specific technical solution adopted by the present invention is as follows:

[0007] An adaptive multi-language translation system, the adaptive multi-language translation system comprising:

[0008] The voice acquisition module is used to acquire the user's audio information and process the audio information to obtain the target audio information;

[0009] A language analysis module is used to perform language analysis on the target audio information, identify the user's language and output language information;

[0010] The semantic recognition module is used to analyze the speech content of the recognized language information, convert the speech content into text, and output the parsed text information;

[0011] A two-way translation module is used to perform forward and reverse translation between the source language and the target language based on the parsed text information, and output the translated text information;

[0012] The adaptive learning module is used to analyze the translated text information and user feedback, and adjust the translation process through optimization algorithms.

[0013] Furthermore, the voice collection module includes:

[0014] An audio acquisition module is used to acquire audio information input by the user in real time;

[0015] The audio processing module is used to preprocess the acquired audio information and perform feature extraction on the preprocessing result to obtain the target audio information.

[0016] Furthermore, the acquired audio information is preprocessed, and feature extraction is performed on the preprocessing result to obtain target audio information including:

[0017] Remove background noise and interference from audio signals, enhance audio signals, and improve the quality of speech signals;

[0018] Normalize the volume level in the audio signal, adjust the channel balance in the audio signal, and remove the echo in the audio signal;

[0019] Analyze the audio signal after echo removal, determine the specific position of each time marker, and identify the key time points in the audio;

[0020] Segmenting the audio signal from which the echo is removed into small segments according to the time markers, each segment including the audio between two time markers;

[0021] Count the total number of time markers per unit time in the audio as the rhythm intensity of the audio;

[0022] When the time-stamp frequency exceeds a predetermined threshold, merging of audio segments is performed;

[0023] A feature extraction algorithm is applied to each audio segment to extract key audio features and obtain representative audio vectors. All audio vectors are merged to construct a complete audio signal feature as the target audio information.

[0024] Further, performing language analysis on the obtained target audio information, identifying the user's language affiliation and outputting language information includes:

[0025] Use feature analysis algorithms to analyze key features of the target audio signal and identify audio attributes that affect the language;

[0026] Combining historical audio data and language knowledge, define the probability of state change between different languages ​​for each key audio feature;

[0027] For each key audio feature, a random value is generated to determine the adjustment measure. If the random value exceeds the preset state change probability, global optimization is performed on the key audio feature. If not, local optimization is performed.

[0028] Check whether the optimization process has reached the predetermined number of iterations. If so, terminate the optimization. If not, continue iterative optimization.

[0029] Output the key audio features after global optimization processing, identify the user's language based on the key audio features, and output the language information.

[0030] Furthermore, the feature analysis algorithm is used to analyze the key features of the target audio signal and identify the audio attributes that affect the language, including:

[0031] Determine the adjustment coefficient and fusion coefficient of the feature analysis algorithm, and set the maximum number of iterations and exploratory searches;

[0032] Based on the key features of the audio signal, the audio is divided into categories and candidate audio features are initialized for each category;

[0033] For each audio category, randomly select from the initialized candidate audio features, and generate new adjusted audio features based on the selected audio features;

[0034] Use adjustment coefficients to fuse different audio features to generate new audio features, compare the newly generated audio features with the current optimal audio features, and retain the audio features with the best performance;

[0035] An exploratory search is performed in the best feature neighborhood of each audio category, and the audio feature with the highest fitness is selected as the audio attribute that affects the language.

[0036] Furthermore, the formula for generating a new adjusted audio feature based on the selected audio feature is:

[0037] Ynew=Ybase+T·(Yrand1-Yrand2);

[0038] In the formula, Y represents the audio feature;

[0039] Y new Represented as new adjusted audio features;

[0040] T represents the adjustment coefficient;

[0041] Y base Representation based audio features;

[0042] Y rand and Y rand Represent randomly selected audio features respectively.

[0043] Furthermore, the semantic recognition module includes:

[0044] The speech-to-text module is used to analyze the speech content of the recognized language information and convert the speech content into text;

[0045] The semantic analysis module is used to perform semantic information analysis on the converted text using a semantic analysis algorithm, identify the meaning of the text content and output it.

[0046] Furthermore, the semantic analysis algorithm is used to analyze the converted text for semantic information, identify the meaning of the text content and output it, including:

[0047] Analyze the text dataset to construct a semantic feature matrix, capture key semantic attributes of the text based on the semantic feature matrix, and identify feature vectors corresponding to zero eigenvalues ​​from the semantic feature matrix;

[0048] Based on each text vector in the text data set, a semantic weight vector is calculated, key feature vectors are extracted from the calculated semantic weight vectors, and spatial analysis is performed on the key feature vectors;

[0049] Construct a semantic transformation matrix based on the key eigenvectors, and analyze the semantic transformation matrix to extract the corresponding key eigenvalues;

[0050] Project a single text vector into the semantic feature space composed of key feature vectors, and use the preset classification criteria to determine the category to which each text belongs in the semantic feature space;

[0051] According to the performance and category of the text in the semantic space, the detailed semantic analysis results of each text are output as the meaning of the text content and output.

[0052] Furthermore, the adaptive learning module includes:

[0053] Feedback analysis module, used to analyze the translated text information and user feedback to determine whether it meets user needs. If so, the final translation result is accepted; if not, translation optimization is performed;

[0054] The optimization module is used to adjust the translation process through optimization algorithms to ensure that the translation results meet the needs of users.

[0055] Furthermore, the translation process is adjusted through optimization algorithms to ensure that the translation results meet the needs of users, including:

[0056] Input various parameters before translation and set translation accuracy constraints;

[0057] Set the number of time periods, maximum number of iterations and total translation time periods during the optimization process;

[0058] Generate the location of the initial translation candidate solution, calculate its fitness value, and record the best translation solution and its fitness value;

[0059] Enter the iterative process, calculate the position of the user feedback center and the sensitive neighborhood radius, and continuously update the position of the feedback center to optimize the translation results;

[0060] Based on user feedback, calculate the translation solution offset caused by guidance adjustment, content adjustment and extension adjustment, and gradually adjust the translated text;

[0061] Execute the perturbation operator to update the candidate translation solutions and their fitness values ​​to ensure that the translation solutions are gradually optimized and approach user needs;

[0062] If the maximum number of iterations is met, the best translation result is output and optimization continues;

[0063] The best translation solution is taken as the translation result, and the final translation version that meets the user's needs is output.

[0064] The beneficial effects of the present invention are:

[0065] 1. The present invention can quickly convert the user's voice information into text and then translate it through automated voice acquisition and processing, which greatly shortens the time required for traditional translation. It also supports the mutual translation function of multiple languages, and can serve a wider user group. By analyzing user feedback and translation results, the translation process is continuously optimized, so that the translation quality will continue to improve with time and data accumulation. At the same time, the forward and reverse translation functions provided by the two-way translation module ensure that users can get good support whether they are inputting or receiving information. Users can switch languages ​​at any time as needed, thereby making communication smoother and promoting the free flow of information and multicultural exchanges.

[0066] 2. The present invention uses a feature analysis algorithm to conduct an in-depth analysis of the target audio signal and identify the audio attributes that have a significant impact on the language. The system can more accurately determine the user's language affiliation. This accurate language identification is the basis for improving translation quality, thereby ensuring the relevance and accuracy of the translation content.

[0067] 3. The present invention uses a semantic analysis algorithm to perform in-depth semantic information analysis on the converted text. The system can not only perform literal translation, but also recognize and understand the deep meaning of the text, thereby helping users to better understand the content of cross-language communication.

[0068] 4. The present invention continuously analyzes user feedback and translation results, adjusts translation strategies and parameters, and makes the translation process more refined and personalized. This continuous learning and optimization mechanism ensures that the translation results increasingly meet the actual needs of users over time, thereby improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0070] Figure 1 The invention is a block diagram of a system for adaptive multi-language translation according to an embodiment of the invention.

[0071] In the figure:

[0072] 1. Speech acquisition module; 2. Language analysis module; 3. Semantic recognition module; 4. Two-way translation module; 5. Adaptive learning module. DETAILED DESCRIPTION

[0073] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0074] According to an embodiment of the present invention, an adaptive multi-language translation system is provided.

[0075] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to an adaptive multi-language translation system according to an embodiment of the present invention, the adaptive multi-language translation system includes:

[0076] The voice acquisition module 1 is used to acquire the user's audio information and process the audio information to obtain the target audio information;

[0077] Specifically, the audio information includes:

[0078] Voice signal: The sound waveform produced by the user's speech or other sound sources.

[0079] Background noise: Various noises in the environment that may be mixed with the main speech signal.

[0080] Non-verbal sounds: such as laughter, crying, or other human or non-human sounds.

[0081] Frequency and amplitude: Basic properties of audio signals, including pitch and volume.

[0082] The target audio information includes:

[0083] Filtered speech signal: A clear speech signal after removing or attenuating background noise and irrelevant sounds.

[0084] Extracted speech features: such as speech pitch, rhythm, speaking speed, etc., which are necessary for language recognition and semantic understanding.

[0085] Segmented speech data: split long audio signals into short segments to facilitate subsequent language analysis and semantic recognition.

[0086] Language analysis module 2, used to perform language analysis on the target audio information, identify the user's language and output language information;

[0087] Specifically, user language attribution refers to the classification of the language used by the user, including:

[0088] Linguistic features of speech: such as intonation, speaking speed, pronunciation characteristics, etc.

[0089] Vocabulary Use: Words and expressions that are common in a particular language.

[0090] Grammatical structure: Different languages ​​have unique grammatical rules.

[0091] Language information refers to specific information about the user's language obtained through analysis by the language analysis module, including: language identification (for example, Chinese, English, Spanish, etc.), dialect or regional differences and language code, etc.

[0092] Semantic recognition module 3 is used to analyze the speech content of the recognized language information, convert the speech content into text, and output the parsed text information;

[0093] A two-way translation module 4 is used to perform forward and reverse translation between the source language and the target language based on the parsed text information, and output the translated text information;

[0094] Specifically, forward translation:

[0095] Definition: The process of translating from a source language to a target language. For example, translating from English to Chinese.

[0096] Purpose: To enable users to understand information or content in a non-native language.

[0097] Back translation:

[0098] Definition: The process of translating from a target language back to the source language. For example, translating English content into Chinese and then back to English.

[0099] Purpose: To verify the accuracy and naturalness of the translation. Usually used for proofreading and ensuring the accuracy of the translated content

[0100] The adaptive learning module 5 is used to analyze the translated text information and user feedback, and adjust the translation process through an optimization algorithm.

[0101] Preferably, the voice collection module 1 includes:

[0102] An audio acquisition module is used to acquire audio information input by the user in real time;

[0103] The audio processing module is used to preprocess the acquired audio information and perform feature extraction on the preprocessing result to obtain the target audio information.

[0104] Preferably, the acquired audio information is preprocessed, and feature extraction is performed on the preprocessing result to obtain the target audio information including:

[0105] Remove background noise and interference from audio signals, enhance audio signals, and improve the quality of speech signals;

[0106] Normalize the volume level in the audio signal, adjust the channel balance in the audio signal, and remove echo in the audio signal;

[0107] Analyze the audio signal after echo removal, determine the specific position of each time marker, and identify the key time points in the audio;

[0108] Segmenting the audio signal from which the echo is removed into small segments according to the time markers, each segment including the audio between two time markers;

[0109] Count the total number of time markers per unit time in the audio as the rhythm intensity of the audio;

[0110] When the time stamp frequency exceeds a predetermined threshold, merging of audio segments is performed;

[0111] A feature extraction algorithm is applied to each audio segment to extract key audio features and obtain representative audio vectors. All audio vectors are merged to construct a complete audio signal feature as the target audio information.

[0112] Specifically, the feature extraction algorithm includes:

[0113] 1) Mel frequency cepstral coefficients:

[0114] Uses: Widely used in speech recognition systems because it can effectively simulate the auditory characteristics of the human ear.

[0115] Principle: First, the audio signal is passed through a series of Mel filters, then the power spectrum of each filter output is taken logarithm, and finally a discrete cosine transform is performed.

[0116] 2) Linear prediction coding coefficients:

[0117] Uses: Common in speech analysis and audio compression, especially for encoding voice information.

[0118] Principle: By predicting the amplitude of the current sample as a linear combination of the amplitudes of previous samples and minimizing the prediction error.

[0119] 3) Cepstral coefficients:

[0120] Application: Used to separate periodic and non-periodic changes of sound signals in audio signal processing.

[0121] Principle: Perform Fourier transform on the signal, take the logarithm, and perform Fourier transform again.

[0122] 4) Spectral Centroid:

[0123] Purpose: Describes the "brightness" of an audio signal and is used for sound analysis of music and speech.

[0124] Principle: Calculate the center of mass distribution in the spectrum, which reflects the brightness of the sound.

[0125] 5) Zero Crossing Rate (ZCR):

[0126] Purpose: A simple way to measure the frequency content of a signal, often used in music beat detection and speech recognition.

[0127] Principle: Calculate the rate at which the waveform in an audio signal passes through zero.

[0128] 6) Energy and energy entropy:

[0129] Purpose: Used for sound activity detection, such as distinguishing between speech and non-speech segments.

[0130] Principle: Energy refers to the sum of the squares of the amplitude of the audio signal, and energy entropy reflects the degree of change in the signal energy distribution.

[0131] Preferably, performing language analysis on the obtained target audio information, identifying the user's language affiliation and outputting language information includes:

[0132] Use feature analysis algorithms to analyze key features of the target audio signal and identify audio attributes that affect the language;

[0133] Combining historical audio data and language knowledge, define the probability of state change between different languages ​​for each key audio feature (i.e., the transition probability in the pollen algorithm);

[0134] For each key audio feature, a random value is generated to determine the adjustment measure. If the random value exceeds the preset state change probability, global optimization is performed on the key audio feature. If not, local optimization is performed.

[0135] Check whether the optimization process has reached the predetermined number of iterations. If so, terminate the optimization. If not, continue iterative optimization.

[0136] Output the key audio features after global optimization processing, identify the user's language based on the key audio features, and output the language information.

[0137] Specifically, the present invention performs language analysis on the target audio information through a pollen algorithm. The idea of ​​the pollen algorithm is derived from the pollination evolution principle of flowering plants in nature, and simulates the entire process of flower pollination. Its basic principle is: first, initialize a number of pollen gametes and other related parameters in the search space, and then, according to the judgment conditions, perform pollination operations on a number of pollen gametes, and select global pollination or local pollination to update the population, repeatedly iterate and optimize, and finally search for the optimal solution to the problem.

[0138] Preferably, the feature analysis algorithm is used to perform key feature analysis on the target audio signal, and the audio attributes that affect the language are identified, including:

[0139] Determine the adjustment coefficient (i.e. scaling factor) and fusion coefficient (i.e. crossover factor) of the feature analysis algorithm, and set the maximum number of iterations and the number of exploratory searches (i.e. chaotic searches);

[0140] Based on the key features of the audio signal, the audio is divided into categories and candidate audio features are initialized for each category;

[0141] For each audio category, randomly select from the initialized candidate audio features, and generate new adjusted audio features based on the selected audio features;

[0142] Use adjustment coefficients to fuse different audio features to generate new audio features, compare the newly generated audio features with the current optimal audio features, and retain the audio features with the best performance;

[0143] An exploratory search is performed in the best feature neighborhood of each audio category, and the audio feature with the highest fitness is selected as the audio attribute that affects the language.

[0144] Specifically, the feature analysis algorithm is to improve the chaotic differential evolution algorithm. In order to accelerate the global convergence speed, the Logistic chaotic mapping is introduced. The chaotic differential evolution algorithm is realized by chaotic search in the solution space centered on the best individual using the ergodicity of chaotic motion.

[0145] Preferably, the formula for generating a new adjusted audio feature based on the selected audio feature is:

[0146] Ynew=Ybase+T·(Yrand1-Yrand2);

[0147] In the formula, Y represents the audio feature;

[0148] Y new Represented as new adjusted audio features;

[0149] T is represented as an adjustment coefficient (ie, a scaling factor) and is between (0, 1).

[0150] Y base Representation based audio features;

[0151] Specifically, Y base It is a specific vector selected from a set of audio feature vectors, which is used as a basic reference point for generating new feature vectors. This baseline vector plays a central role in the differential evolution algorithm, because the generation of new vectors is based on this baseline vector and by introducing differences between other vectors.

[0152] Y rand1 and Y rand2 Respectively represent randomly selected audio features (random positive integers).

[0153] Preferably, the semantic recognition module 3 includes:

[0154] The speech-to-text module is used to analyze the speech content of the recognized language information and convert the speech content into text;

[0155] The semantic analysis module is used to perform semantic information analysis on the converted text using a semantic analysis algorithm, identify the meaning of the text content and output it.

[0156] Preferably, using a semantic analysis algorithm to perform semantic information analysis on the converted text, identifying the meaning of the text content and outputting it includes:

[0157] Analyze the text dataset to construct a semantic feature matrix, capture key semantic attributes of the text based on the semantic feature matrix, and identify feature vectors corresponding to zero eigenvalues ​​from the semantic feature matrix;

[0158] Based on each text vector in the text data set, a semantic weight vector is calculated, key feature vectors are extracted from the calculated semantic weight vectors, and spatial analysis is performed on the key feature vectors;

[0159] Construct a semantic transformation matrix based on the key eigenvectors, and analyze the semantic transformation matrix to extract the corresponding key eigenvalues;

[0160] Project a single text vector into the semantic feature space composed of key feature vectors, and use the preset classification criteria to determine the category to which each text belongs in the semantic feature space;

[0161] According to the performance and category of the text in the semantic space, the detailed semantic analysis results of each text are output as the meaning of the text content and output.

[0162] Specifically, the semantic analysis algorithm is the improved linear discriminant subspace pattern recognition algorithm (I-VLDNS algorithm), which is an improvement on the traditional linear discriminant analysis (LDA). It is mainly used in the semantic analysis of text data, and improves the accuracy of pattern recognition and classification by deeply understanding the structure and semantics of the text. It includes:

[0163] 1) Constructing semantic feature matrix:

[0164] Process: First, the algorithm analyzes the text dataset and extracts the key semantic attributes of the text. This step usually involves text preprocessing, such as word segmentation, stop word removal, and feature calculation such as word frequency, TF-IDF, etc.

[0165] Purpose: To construct a semantic feature matrix that contains key information to represent text data. This matrix is ​​the basis for subsequent analysis.

[0166] 2) Calculate and extract key feature vectors:

[0167] Process: Based on the above semantic feature matrix, a semantic weight vector is calculated, which usually involves dimensionality reduction techniques such as PCA or LDA to extract the main variables of the data. Key feature vectors are further extracted from these semantic weight vectors.

[0168] Purpose: Key feature vectors can capture the most significant semantic differences in text datasets and are used to distinguish different categories or topics.

[0169] 3) Construct and analyze the semantic transformation matrix:

[0170] Process: Based on the key feature vectors, a semantic transformation matrix is ​​constructed. This matrix is ​​used to transform the text from the original feature space to a new semantic space, where different semantic categories are easier to distinguish.

[0171] Purpose: Through this transformation, the semantic structure of the text is more clearly presented, facilitating more precise classification and interpretation.

[0172] 4) Text classification in semantic feature space:

[0173] Process: Project a single text vector into a semantic feature space composed of key feature vectors. In this space, the position of each text more clearly represents its semantic attributes. Use the preset classification criteria (such as nearest neighbor, SVM, etc.) to determine the category to which each text belongs.

[0174] Purpose: This step allows the algorithm to automatically classify the text, identify its topic or sentiment, etc., based on its representation in the semantic feature space.

[0175] 5) Output detailed semantic analysis results:

[0176] Process: Finally, based on the text’s performance and category in the semantic space, the algorithm outputs detailed semantic analysis results for each text. This includes the text’s key themes, sentiment tendencies, style features, etc.

[0177] Purpose: Provide in-depth text interpretation to assist in decision making, content recommendation or other applications.

[0178] Preferably, the adaptive learning module 5 comprises:

[0179] Feedback analysis module, used to analyze the translated text information and user feedback to determine whether it meets user needs. If so, the final translation result is accepted; if not, translation optimization is performed;

[0180] The optimization module is used to adjust the translation process through optimization algorithms to ensure that the translation results meet the needs of users.

[0181] Preferably, the translation process is adjusted by optimizing the algorithm to ensure that the translation result meets the user's needs including:

[0182] Enter various parameters before translation (including the original text entered by the user, translation preferences, context, etc.), and set translation accuracy constraints;

[0183] Set the number of time periods, maximum number of iterations and total translation time periods during the optimization process;

[0184] Generate the location of the initial translation candidate solution (i.e., the translation text), calculate its fitness value (i.e., the degree of matching with user requirements), and record the best translation solution and its fitness value;

[0185] Enter the iterative process, calculate the position of the user feedback center and the sensitive neighborhood radius, and continuously update the position of the feedback center to optimize the translation results;

[0186] Based on user feedback, calculate the translation solution offset caused by guidance adjustment, content adjustment and extension adjustment, and gradually adjust the translated text;

[0187] Execute the perturbation operator to update the candidate translation solutions and their fitness values ​​to ensure that the translation solutions are gradually optimized and approach user needs;

[0188] If the maximum number of iterations is met or the translation solution satisfaction reaches the standard, the best translation result is output and optimization continues;

[0189] The best translation solution is used as the translation result or optimized again according to the user's further feedback to output the final translation version that meets the user's needs.

[0190] Specifically, the optimization algorithm is the krill swarm algorithm, the basic principle of which is to simulate the clustering behavior of krill in nature to solve the optimization problem. This algorithm consists of three main behaviors: induced movement, foraging behavior, and random diffusion. The following is a detailed description of each behavior:

[0191] 1) Induced movement:

[0192] Purpose: To simulate the behavior of krill moving in groups by influencing each other.

[0193] Principle: The position update of each krill individual is affected by the current optimal krill position and the center of the group. This motion component takes into account the local optimal solution and the overall goal of the group, helping the algorithm maintain the diversity and concentration of the search.

[0194] 2) Foraging behavior:

[0195] Purpose: To reflect the behavior of krill in finding food, that is, to find the optimal solution in the optimization problem.

[0196] Principle: The foraging behavior of krill consists of two parts: directed movement toward the food source (i.e., the currently known best solution) and movement in the gradient direction (i.e., the direction with the largest gradient of the fitness function in the solution space). This behavior enables krill to move toward better solution areas.

[0197] 3) Random Diffusion:

[0198] Purpose: To simulate the random swimming of krill in water and increase the ability to explore unknown areas.

[0199] Principle: Each position update of krill includes a random component, which increases the ability of the algorithm to jump out of the local optimal solution, while providing additional exploration capabilities to avoid premature convergence.

[0200] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An adaptive multilingual translation system, characterized in that: The adaptive multilingual translation system includes: The voice acquisition module is used to acquire the user's audio information and process the audio information to obtain the target audio information; A language analysis module is used to perform language analysis on the target audio information, identify the user's language and output language information; The semantic recognition module is used to analyze the speech content of the recognized language information, convert the speech content into text, and output the parsed text information; A two-way translation module is used to perform forward and reverse translation between the source language and the target language based on the parsed text information, and output the translated text information; The adaptive learning module is used to analyze the translated text information and user feedback, and adjust the translation process through optimization algorithms.

2. The adaptive multi-language translation system according to claim 1, characterized in that: The voice collection module comprises: An audio acquisition module is used to acquire audio information input by the user in real time; The audio processing module is used to preprocess the acquired audio information and perform feature extraction on the preprocessing result to obtain the target audio information.

3. The adaptive multi-language translation system according to claim 2, characterized in that: The obtained audio information is preprocessed, and features are extracted from the preprocessing results to obtain target audio information, including: Remove background noise and interference from audio signals, enhance audio signals, and improve the quality of speech signals; Normalize the volume level in the audio signal, adjust the channel balance in the audio signal, and remove echo in the audio signal; Analyze the audio signal after echo removal, determine the specific position of each time marker, and identify the key time points in the audio; Segmenting the audio signal from which the echo is removed into small segments according to the time markers, each segment including the audio between two time markers; Count the total number of time markers per unit time in the audio as the rhythm intensity of the audio; When the time stamp frequency exceeds a predetermined threshold, merging of audio segments is performed; A feature extraction algorithm is applied to each audio segment to extract key audio features and obtain representative audio vectors. All audio vectors are merged to construct a complete audio signal feature as the target audio information.

4. The adaptive multi-language translation system according to claim 1, characterized in that: The performing language analysis on the obtained target audio information, identifying the user's language affiliation and outputting the language information comprises: Use feature analysis algorithms to analyze key features of the target audio signal and identify audio attributes that affect the language; Combining historical audio data and language knowledge, define the probability of state change between different languages ​​for each key audio feature; For each key audio feature, a random value is generated to determine the adjustment measure. If the random value exceeds the preset state change probability, global optimization is performed on the key audio feature. If not, local optimization is performed. Check whether the optimization process has reached the predetermined number of iterations. If so, terminate the optimization. If not, continue iterative optimization. Output the key audio features after global optimization processing, identify the user's language based on the key audio features, and output the language information.

5. The adaptive multi-language translation system according to claim 4, characterized in that: The use of a feature analysis algorithm to perform key feature analysis on the target audio signal and identify the audio attributes that affect the language includes: Determine the adjustment coefficient and fusion coefficient of the feature analysis algorithm, and set the maximum number of iterations and exploratory searches; Based on the key features of the audio signal, the audio is divided into categories and candidate audio features are initialized for each category; For each audio category, randomly select from the initialized candidate audio features, and generate new adjusted audio features based on the selected audio features; Use adjustment coefficients to fuse different audio features to generate new audio features, compare the newly generated audio features with the current optimal audio features, and retain the audio features with the best performance; An exploratory search is performed in the best feature neighborhood of each audio category, and the audio feature with the highest fitness is selected as the audio attribute that affects the language.

6. The adaptive multi-language translation system according to claim 5, characterized in that: The formula for generating a new adjusted audio feature based on the selected audio feature is: Ynew=Ybase+T·(Yrand1-Yrand2); In the formula, Y represents the audio feature; Y new Represented as new adjusted audio features; T represents the adjustment coefficient; Y base Representation based audio features; Y rand1 and Y rand2 Represent randomly selected audio features respectively.

7. The adaptive multi-language translation system according to claim 1, characterized in that: The semantic recognition module comprises: The speech-to-text module is used to analyze the speech content of the recognized language information and convert the speech content into text; The semantic analysis module is used to perform semantic information analysis on the converted text using a semantic analysis algorithm, identify the meaning of the text content and output it.

8. The adaptive multi-language translation system according to claim 7, characterized in that: The method of using a semantic analysis algorithm to analyze the converted text for semantic information, identifying the meaning of the text content and outputting the content includes: Analyze the text dataset to construct a semantic feature matrix, capture key semantic attributes of the text based on the semantic feature matrix, and identify feature vectors corresponding to zero eigenvalues ​​from the semantic feature matrix; Based on each text vector in the text data set, a semantic weight vector is calculated, key feature vectors are extracted from the calculated semantic weight vectors, and spatial analysis is performed on the key feature vectors; Construct a semantic transformation matrix based on the key eigenvectors, and analyze the semantic transformation matrix to extract the corresponding key eigenvalues; Project a single text vector into the semantic feature space composed of key feature vectors, and use the preset classification criteria to determine the category to which each text belongs in the semantic feature space; According to the performance and category of the text in the semantic space, the detailed semantic analysis results of each text are output as the meaning of the text content and output.

9. The adaptive multi-language translation system according to claim 1, characterized in that: The adaptive learning module includes: Feedback analysis module, used to analyze the translated text information and user feedback to determine whether it meets user needs. If so, the final translation result is accepted; if not, translation optimization is performed; The optimization module is used to adjust the translation process through optimization algorithms to ensure that the translation results meet the needs of users.

10. The adaptive multi-language translation system according to claim 9, characterized in that: The adjustment of the translation process by optimizing the algorithm to ensure that the translation result meets the needs of the user includes: Input various parameters before translation and set translation accuracy constraints; Set the number of time periods, maximum number of iterations and total translation time periods during the optimization process; Generate the location of the initial translation candidate solution, calculate its fitness value, and record the best translation solution and its fitness value; Enter the iterative process, calculate the position of the user feedback center and the sensitive neighborhood radius, and continuously update the position of the feedback center to optimize the translation results; Based on user feedback, calculate the translation solution offset caused by guidance adjustment, content adjustment and extension adjustment, and gradually adjust the translated text; Execute the perturbation operator to update the candidate translation solutions and their fitness values ​​to ensure that the translation solutions are gradually optimized and approach user needs; If the maximum number of iterations is met, the best translation result is output and optimization continues; The best translation solution is taken as the translation result, and the final translation version that meets the user's needs is output.

Images