Agricultural information handling method and system based on virtual digital human and AI

Abstract

The invention discloses an agricultural information handling method and system based on a virtual digital human and an AI, and relates to the technical field of agricultural information handling, and the method comprises the following steps: firstly, capturing user voice input data in real time, collecting user voice parameters, and providing original data support for subsequent analysis and processing; and summarizing the obtained voice data parameters into an analysis set, and preprocessing the collected voice data. According to the method, the voice of the user is captured in real time, the change of the voice speed is dynamically analyzed, the voice speed is intelligently evaluated by using the deep learning model, adaptive adjustment is performed for urgent voice, the sensitivity is reduced, the processing time is prolonged, and the accuracy of voice recognition is ensured. The method is especially suitable for agricultural information handling, helps farmers to obtain accurate agricultural production suggestions, reduces misrecognition, decision errors, resource waste and production loss, and improves agricultural benefits and service quality.

Description

Technical Field

[0001] The present invention relates to the field of agricultural information management technology, and in particular to an agricultural information management method and system based on virtual digital humans and AI. Background Art

[0002] Agricultural information processing based on virtual digital humans and AI refers to the use of virtual digital humans (i.e., digital avatars or "virtual assistants") combined with artificial intelligence technology to provide intelligent information processing and service systems for the agricultural sector. Through AI technologies such as speech recognition, natural language processing, and semantic understanding, the system can answer farmers' or users' agricultural-related questions in real time and provide services such as agricultural production guidance, market analysis, and policy consulting. Virtual digital humans can not only interact with users in a friendly and visual manner, but also enhance the effectiveness of information transmission through digital movements and expressions, thereby improving the efficiency and interactivity of agricultural information dissemination. This system can replace traditional manual services or paper materials, providing faster, more accurate, and easier-to-understand agricultural information, helping farmers better master agricultural knowledge, obtain policy information, and improve the efficiency of agricultural production and management.

[0003] The existing technology has the following shortcomings: In the existing technology, when processing user voice, speech recognition systems often make recognition errors if the speech is spoken too quickly. Such errors may cause the system to fail to accurately understand the user's questions or needs, leading to misunderstandings. Especially in the process of processing agricultural information, users' questions may involve professional agricultural terminology or complex production instructions. If the system fails to correctly recognize these contents, it may provide inaccurate agricultural production suggestions or those that do not meet the user's needs. This not only affects the effectiveness of information transmission, but may also cause farmers to make decisions based on incorrect guidance, resulting in wasted resources or losses in crop production. For example, incorrect fertilization or irrigation recommendations may hinder crop growth and may even affect the overall benefits of agricultural production.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for agricultural information management based on virtual digital humans and AI. This method captures user speech in real time and dynamically analyzes speech rate changes. It utilizes a deep learning model to intelligently assess speech rate, adaptively adjusts for rapid speech, reduces sensitivity, and prolongs processing time to ensure accurate speech recognition. This method is particularly suitable for agricultural information management, helping farmers obtain accurate agricultural production recommendations, reducing misidentification, decision-making errors, resource waste, and production losses, improving agricultural efficiency and service quality, and addressing the aforementioned background technology issues.

[0006] In order to achieve the above objectives, the present invention provides the following technical solution: a method for handling agricultural information based on virtual digital humans and AI, comprising the following steps:

[0007] First, it captures user voice input data in real time, collects user voice parameters, and provides raw data support for subsequent analysis and processing;

[0008] Aggregate the acquired voice data parameters into an analysis set and pre-process the collected voice data;

[0009] Extract key features reflecting speech rate changes from preprocessed speech data and dynamically analyze the extracted key features within the detection window to quantify the user's speech rate changes.

[0010] The quantified features are input into a pre-trained deep learning model, and the pre-trained deep learning model is used to intelligently evaluate the current user's speaking speed.

[0011] When the deep learning model assesses that the user is speaking rapidly, it makes adaptive adjustments based on the user's voice input frequency and speaking speed to reduce sensitivity to rapid speech, making it more tolerant of rapid speech input and reducing the probability of misrecognition. In addition, the processing time of the voice input is extended according to the rapidity of the speaking speed, providing more processing space for the speech recognition process, thereby ensuring that even rapid speech can be accurately parsed.

[0012] Preferably, the specific steps of aggregating the acquired voice data parameters into an analysis set and performing preprocessing include:

[0013] First, collect user voice data and extract multiple key parameters;

[0014] Then, the parameters are aggregated into an analysis set in chronological order to form structured data for subsequent analysis;

[0015] Next, the speech data is preprocessed, including noise removal, signal normalization, and removal of useless data to ensure the clarity and consistency of the audio data;

[0016] During pre-processing, automatic speech segmentation is performed to effectively handle long speech inputs.

[0017] The preprocessed data will provide high-quality basic data for subsequent feature extraction, speech rate analysis, and deep learning model input.

[0018] Preferably, key features reflecting changes in speech rate are extracted from the preprocessed speech data, the extracted features including changes in speech volume per unit time and the unevenness of duration between words. Within the detection window, the changes in speech volume per unit time and the unevenness of duration between words are analyzed and processed to generate a volume fluctuation amplitude reference value and a speech duration irregularity reference value, respectively. The volume fluctuation amplitude reference value and the speech duration irregularity reference value are used to quantify the degree of speech urgency and rhythm instability.

[0019] Preferably, the specific steps of analyzing the change of the voice volume within a unit time in the detection window to generate the volume fluctuation amplitude reference value are as follows:

[0020] First, perform frame processing on the speech signal, divide the speech signal into multiple small time windows, and calculate the volume change in each window. The calculation expression is as follows:

[0021]

[0022] Where x i，j is the audio amplitude of the jth sampling point in the i-th frame, n is the number of sampling points in the current frame, V i is the instantaneous fluctuation amplitude of volume change;

[0023] Get the instantaneous fluctuation amplitude V of the volume change i Finally, the volume fluctuation amplitude is quantified and a volume fluctuation amplitude reference value is generated. The quantization formula is as follows:

[0024]

[0025] Where V index is the volume fluctuation amplitude reference value, T is the total duration of the speech signal, N is the total number of signal frames, max(V) is the maximum value of the volume fluctuation amplitude, and α is the weighting coefficient used to control the degree of influence of the volume fluctuation amplitude on the final volume fluctuation reference value.

[0026] Preferably, the specific steps of analyzing the unevenness of the duration between words in the detection window to generate the reference value of speech duration irregularity are as follows:

[0027] In this stage, we first extract the time interval between words in the speech signal. By framing and phoneme analysis of the speech signal, we accurately detect the start and end time of each word, and then calculate the duration difference between words. For each pair of words, we calculate the duration difference, that is, the time interval between words. The calculation expression is as follows:

[0028] ΔT q =|T end，q -T start，q+1 |

[0029] Where, ΔT q is the time difference between the qth word pair, that is, the duration time interval between two adjacent words, T end，q is the ending time of the qth word, T start，q+1 is the starting time of the q+1th word;

[0030] After calculating the time difference ΔT between word pairs q Finally, we further quantify the fluctuation range of the difference and adopt the weighted reference value model to quantify its volatility through the nonlinear function of the interval time difference between word pairs. The formula is as follows:

[0031]

[0032] Where W q is the reference value of the fluctuation amplitude of the qth word pair, max(ΔT) is the maximum time difference between all word pairs in the entire speech signal, count q is the frequency of the qth word pair in the speech signal, θ and β are adjustment coefficients, θ is used to control the intensity of the influence of the time difference on the reference value of the fluctuation amplitude, and β is used to control the influence of the word pair frequency on the reference value of the fluctuation amplitude;

[0033] Finally, by summarizing the fluctuation amplitudes of all word pairs, a reference value of speech duration irregularity is generated. The generation formula is as follows:

[0034]

[0035] Where R duration is the reference value of speech duration irregularity, M is the total number of word pairs in the speech signal, exp(-γ(ΔT q -ρ)) is the reference value attenuation function, γ is the reference value attenuation coefficient, which controls the steepness of the reference value function, and ρ is the threshold of the time difference.

[0036] Preferably, the analyzed volume fluctuation amplitude reference value and speech duration irregularity reference value are input into a pre-trained deep learning model, and a speech rate variation coefficient is generated by the deep learning model, and the current user's speech rate is intelligently evaluated by the speech rate variation coefficient.

[0037] Preferably, the speech rate variation coefficient generated when the pre-trained deep learning model is used to intelligently evaluate the current user's speech rate is compared with a pre-set speech rate variation coefficient reference threshold to divide the current user's speech rate. The specific steps are as follows:

[0038] If the speech rate variation coefficient is greater than the preset speech rate variation coefficient reference threshold, the current user speech rate is classified as rapid speech; if the speech rate variation coefficient is less than or equal to the preset speech rate variation coefficient reference threshold, the current user speech rate is classified as slow speech.

[0039] Preferably, when the deep learning model assesses that the user's speech speed is rapid, adaptive adjustments are made based on the user's speech input frequency and speech speed to reduce sensitivity to rapid speech. In addition, the specific steps of extending the processing time of the speech input according to the rapidity of the speech speed are as follows:

[0040] When the deep learning model determines that the user is speaking rapidly, it will reduce its sensitivity to voice input and reduce misrecognition caused by rapid speech. The calculation expression is as follows:

[0041]

[0042] Where, α new is the adjusted sensitivity, α base is the default sensitivity, S peech is the speech rate variation coefficient, S ref is the reference threshold of speech rate variation coefficient, S max is the maximum speech rate variation coefficient, ω is a nonlinear adjustment parameter that controls the degree of the inverse relationship between the speech rate variation coefficient and sensitivity, and δ is the amplification factor of the sensitivity adjustment, which can provide a stronger adjustment for rapid speech rate;

[0043] When a rapid speech rate is detected, the processing time of the speech input is dynamically extended according to the rapidity of the speech rate. The extended processing time is proportional to the speech rate change coefficient S. peech There is a nonlinear relationship between them and the reference threshold S of the speech rate variation coefficient. ref The calculation formula is as follows:

[0044]

[0045] Where, T ext is the extended processing time, T base is the default speech processing time, κ and ∈ are nonlinear adjustment coefficients, κ controls the extension of processing time when the speech rate is fast, ∈ controls the nonlinear relationship between speech rate change and processing time extension, κ and ∈ are both greater than 1, λ is the smoothing coefficient;

[0046] By adjusting the sensitivity of the voice input and extending the processing time, the voice recognition process is automatically optimized when rapid speech is detected. peech and the speech rate variation coefficient reference threshold S refDynamic adjustments make the system more tolerant to fast voice input, while also ensuring longer voice processing time to improve voice recognition accuracy. The overall effect of the comprehensive adjustments is quantified by the following formula:

[0047]

[0048] Where RecognitionAccuracy is the accuracy of speech recognition, is the impact index of rapid speaking rate on recognition accuracy, σ is the amplification coefficient of the impact of speech rate change on recognition accuracy, and ρ is the smoothing coefficient.

[0049] Preferably, an agricultural information management system based on virtual digital humans and AI includes a voice data acquisition module, a voice data preprocessing module, a speech rate feature extraction and dynamic analysis module, a deep learning evaluation module, and an adaptive adjustment and processing time extension module:

[0050] The voice data acquisition module first captures user voice input data in real time, collects user voice parameters, and provides raw data support for subsequent analysis and processing;

[0051] The voice data preprocessing module aggregates the acquired voice data parameters into an analysis set and preprocesses the collected voice data;

[0052] The speech rate feature extraction and dynamic analysis module extracts key features reflecting speech rate changes from preprocessed speech data and dynamically analyzes the extracted key features within the detection window to quantify the user's speech rate changes.

[0053] The deep learning evaluation module inputs the quantified features into a pre-trained deep learning model and uses the pre-trained deep learning model to intelligently evaluate the current user's speaking speed;

[0054] The adaptive adjustment and processing time extension module makes adaptive adjustments based on the user's voice input frequency and speed to reduce sensitivity to rapid speech when the deep learning model assesses that the user is speaking rapidly, making it more tolerant of rapid speech input and reducing the probability of misrecognition. In addition, the processing time of voice input is extended according to the rapidity of the speech rate, providing more processing space for the speech recognition process, thereby ensuring that even rapid speech can be accurately parsed.

[0055] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0056] The present invention effectively solves the problem of speech recognition errors caused by rapid speech speed by capturing user voice in real time and dynamically analyzing changes in speech speed. It is particularly suitable for scenarios involving professional terminology and complex guidance content, such as agricultural information processing. The method uses a deep learning model to intelligently evaluate speech speed, and makes adaptive adjustments based on the user's voice input frequency and the rapidity of the speech speed, reducing sensitivity to rapid speech while extending the processing time of voice input. Through these adjustments, the present invention ensures that the system can accurately understand user needs when faced with fast speech, reduces the probability of misrecognition, and improves recognition accuracy. Ultimately, farmers can obtain more accurate agricultural production recommendations, avoid decision-making errors caused by misleading information, reduce resource waste and agricultural production losses, and thus improve agricultural production efficiency and service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0058] Figure 1 This is a method flow chart of an agricultural information management method based on virtual digital humans and AI in the present invention.

[0059] Figure 2 This is a module schematic diagram of an agricultural information management system based on virtual digital humans and AI in the present invention. DETAILED DESCRIPTION

[0060] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0061] The present invention provides Figure 1 The agricultural information processing method based on virtual digital human and AI shown includes the following steps:

[0062] First, user voice input data is captured in real time, and user voice parameters (such as volume, frequency, speaking speed, pitch, syllable duration, etc.) are collected to provide raw data support for subsequent analysis and processing;

[0063] First, the speech recognition system captures the user's voice signal through a microphone or voice input device. After preliminary sampling and signal processing, this signal is converted into digital data. This data includes but is not limited to voice features such as volume, speaking rate, and intonation, providing essential information about the user's voice. Through this process, the system can monitor changes in the user's voice in real time, laying the foundation for subsequent voice feature extraction and processing. For example, changes in volume and speaking rate can serve as a dynamic indicator of the user's voice, making it particularly important to capture these changes in a timely manner, especially when speaking too quickly.

[0064] Aggregate the acquired voice data parameters into an analysis set and pre-process the collected voice data;

[0065] The specific steps for aggregating the acquired voice data parameters into an analysis set and preprocessing them include: First, the system collects the user's voice data and extracts multiple key parameters, such as volume, speaking rate, frequency, intonation, and duration of the voice. These parameters are then aggregated into an analysis set in chronological order to form structured data for subsequent analysis. Next, the voice data is preprocessed, including noise removal, signal normalization, and removal of useless data, to ensure the clarity and consistency of the audio data. During the preprocessing process, the system may perform automated voice segmentation to more efficiently process long voice inputs. The preprocessed data will provide high-quality basic data for subsequent feature extraction, speaking rate analysis, and deep learning model input.

[0066] Establishing an analysis set provides structured storage for voice data, facilitating batch processing and comparative analysis. This lays the foundation for subsequent feature extraction and deep learning model training, ensuring the system's efficiency and consistency in processing diverse user voices. This data includes information such as the user's pronunciation rate, volume fluctuations, and pitch changes. Constructing an analysis set helps the system aggregate and categorize all input, providing a manageable structure for subsequent in-depth analysis. All voice input data is stored and labeled for easy comparison and processing.

[0067] After acquiring the raw voice data, the system performs data preprocessing, including denoising, signal enhancement, normalization, and redundancy removal. This step is crucial, as voice data is often accompanied by interference from background noise, echoes, and other factors. Directly using unprocessed data can lead to recognition errors. Denoising techniques remove ambient noise while maintaining the clarity of the voice signal. Signal enhancement improves signal quality, making voice features more prominent. Furthermore, normalization ensures that voice data from different environments meets the same standards, providing a consistent data source for subsequent feature extraction.

[0068] Extract key features reflecting speech rate changes from preprocessed speech data and dynamically analyze the extracted key features within the detection window to quantify the user's speech rate changes.

[0069] Key features reflecting changes in speech rate are extracted from the preprocessed speech data. The extracted features include the change in speech volume per unit time and the unevenness of the duration between words. Within the detection window, the change in speech volume per unit time and the unevenness of the duration between words are analyzed and processed to generate reference values ​​for volume fluctuation amplitude and speech duration irregularity, respectively. The volume fluctuation amplitude reference value and speech duration irregularity reference value are used to quantify the degree of speech urgency and rhythm instability.

[0070] Rapid changes in voice volume within a unit of time often indicate that the user is speaking too quickly. This is because fast speech often causes erratic rhythm and large fluctuations in volume. Specifically, at a normal speaking rate, voice volume fluctuations are generally relatively stable. However, when the user speaks faster, nervousness or urgency can cause frequent and large fluctuations in volume within a short period of time. These fluctuations are not only caused by airflow changes during pronunciation, but also by the speaker's breathing rate and vocal control. At fast speaking rates, users may neglect appropriate pauses, resulting in a loss of rhythm and more dramatic volume fluctuations. In particular, when speaking too quickly, the movement of the vocal organs accelerates, increasing muscle tension during articulation, leading to more frequent volume fluctuations. These rapid volume fluctuations can serve as a key feature to identify rapid speech, helping the speech recognition system determine whether the user is speaking too quickly. By analyzing the amplitude and frequency of volume fluctuations, the system can effectively identify rapidly spoken input and make appropriate adjustments to avoid misrecognition or information loss caused by excessive speech.

[0071] The specific steps for analyzing the change in speech volume per unit time within the detection window to generate a reference value for the volume fluctuation amplitude are as follows:

[0072] First, perform frame processing on the speech signal, divide the speech signal into multiple small time windows, and calculate the volume change in each window. The calculation expression is as follows:

[0073]

[0074] Where x i，j is the audio amplitude of the jth sampling point in the i-th frame, n is the number of sampling points in the current frame, V i is the instantaneous fluctuation amplitude of volume change;

[0075] The amplitude difference between consecutive sampling points is calculated to capture the magnitude of volume fluctuations. Larger values ​​indicate more dramatic volume fluctuations, often associated with rapid speech. By framing and calculating instantaneous changes, the system can capture subtle differences in volume changes, providing a basis for determining whether speech is rapid or not.

[0076] Get the instantaneous fluctuation amplitude V of the volume change i Finally, the volume fluctuation amplitude is quantified and a volume fluctuation amplitude reference value is generated. The quantization formula is as follows:

[0077]

[0078] Where V index is the volume fluctuation amplitude reference value, T is the total duration of the speech signal, N is the total number of signal frames, max(V) is the maximum value of the volume fluctuation amplitude, and α is the weighting coefficient used to control the degree of influence of the volume fluctuation amplitude on the final volume fluctuation reference value.

[0079] By normalizing the volume fluctuation amplitude of each frame The volume fluctuations are then weighted according to the α setting to generate a final volume fluctuation reference value. This reference value effectively quantifies the rapidity of the speech input based on the weight of the time interval and volume fluctuations. A larger reference value indicates more severe volume fluctuations and a faster speaking rate.

[0080] A larger volume fluctuation amplitude reference value, generated by analyzing the change in voice volume over a unit time within the detection window, generally indicates that the user is speaking too quickly. This is because the volume fluctuation amplitude reference value measures the change in voice volume over a unit time. When a user speaks faster, the rhythm of the speech tends to become erratic, and volume fluctuations become more frequent and dramatic. When speaking quickly, the user often doesn't have enough time for steady breathing and regulation, resulting in large volume fluctuations within a short period of time. On the other hand, when speaking at a slower rate, the rhythm of the speech is more stable, and the volume fluctuations are smaller. Therefore, a larger volume fluctuation amplitude reference value indicates a faster speaking rate, and vice versa.

[0081] Increased unevenness in the duration between words often indicates that the user is speaking too quickly. This is because when a user speaks quickly, pauses between words are often short, or even nonexistent. Normally, there is a certain amount of time between each word or phrase in speech input, providing clear speech boundaries. However, when speaking too quickly, the user rapidly pronounces the next word with little or no pause. In this case, the time differences between words become irregular, leading to increased unevenness in speech duration. In other words, rapid speaking causes the intervals between each word to become unstable, making it impossible to maintain a normal sense of rhythm and speech fluency, thus affecting the clarity and intelligibility of the speech. Therefore, increased unevenness in the duration between words is a typical sign of rapid speech, indicating that the user's speech rhythm is unstable, and the speech recognition system needs to adjust to this to ensure more accurate understanding of the user's intent and reduce recognition errors.

[0082] The specific steps for analyzing the unevenness of the duration between words in the detection window to generate a reference value for speech duration irregularity are as follows:

[0083] In this stage, we first extract the time interval between words in the speech signal. By framing and phoneme analysis of the speech signal, we accurately detect the start and end time of each word, and then calculate the duration difference between words. For each pair of words, we calculate the duration difference, that is, the time interval between words. The calculation expression is as follows:

[0084] ΔT q =|T end，q -T start，q+1 |

[0085] , where ΔT q is the time difference between the qth word pair, that is, the duration time interval between two adjacent words, T end，q is the ending time of the qth word, T start，q+1 is the starting time of the q+1th word;

[0086] Through these steps, the system can capture the time difference between each pair of words, providing basic data for subsequent analysis. This difference directly reflects the smoothness of speech rate. When the pauses are long or irregular, it indicates that the speech rate may be rapid.

[0087] After calculating the time difference ΔT between word pairs q Finally, we further quantify the fluctuation range of the difference and adopt the weighted reference value model to quantify its volatility through the nonlinear function of the interval time difference between word pairs. The formula is as follows:

[0088]

[0089] Where W q is the reference value of the fluctuation amplitude of the qth word pair, max(ΔT) is the maximum time difference between all word pairs in the entire speech signal, count q is the frequency of the qth word pair in the speech signal, θ and β are adjustment coefficients, θ is used to control the intensity of the influence of the time difference on the reference value of the fluctuation amplitude, and β is used to control the influence of the word pair frequency on the reference value of the fluctuation amplitude;

[0090] This weighted reference allows the system to more accurately reflect the unevenness of word spacing. If the time differences between certain word pairs are too large and they occur frequently, their contribution to the fluctuation amplitude will be higher, which helps to reveal the rapidity of speech.

[0091] Finally, by summarizing the fluctuation amplitudes of all word pairs, a reference value of speech duration irregularity is generated. The generation formula is as follows:

[0092]

[0093] Where R duration is the reference value of speech duration irregularity, M is the total number of word pairs in the speech signal, exp(-γ(ΔT q -ρ)) is the reference value attenuation function, γ is the reference value attenuation coefficient, which controls the steepness of the reference value function, and ρ is the threshold of time difference, which serves as an offset of the reference value function to help the system identify time differences that are beyond the normal range.

[0094] Generated speech duration irregularity reference value R duration This value provides a comprehensive measure of uneven spacing between words and rapid speech. A high reference value indicates an unstable speech rhythm and irregular spacing between words, suggesting the user is speaking rapidly. Using this reference value, the system can effectively detect and quantify the degree of rapid speech, enabling more accurate judgment and adjustment of speech recognition.

[0095] A larger Speech Duration Irregularity Reference Value, generated by analyzing the uneven duration between words within the detection window, generally indicates a faster speaking rate. This is because a faster speaking rate results in shorter, or even almost no, pauses between words, resulting in inconsistent or irregular durations between each word. Conversely, if the speaking rate is slower, the pauses between words are typically more uniform, and the speech rhythm is more stable, resulting in a smaller Speech Duration Irregularity Reference Value.

[0096] The quantified features are input into a pre-trained deep learning model, and the pre-trained deep learning model is used to intelligently evaluate the current user's speaking speed.

[0097] The analyzed volume fluctuation amplitude reference value and speech duration irregularity reference value are input into a pre-trained deep learning model, and a speech rate variation coefficient is generated by the deep learning model. The current user's speech rate is intelligently evaluated by the speech rate variation coefficient.

[0098] A pre-trained machine learning model is one that is trained using a large amount of labeled speech data before the system is deployed. After multiple rounds of optimization and adjustment, the model is ensured to accurately and intelligently analyze speech data in actual applications. The training process of this model usually involves supervised learning methods, which contain a large number of speech data samples that have been manually labeled into different speech speed categories, such as "normal speaking speed" and "rapid speaking speed." During the training process, the model continuously adjusts its internal parameters to learn the relationship between different speech features (such as volume fluctuations, pitch changes, and the duration of pauses between words) and speech speed changes. Once the model training is complete, it has the ability to process new input data, automatically determine speech speed changes based on the features extracted from the input data, and generate a speech speed change coefficient.

[0099] The core purpose of this machine learning model is to associate complex speech features with patterns of speech rate changes, thereby automatically assessing whether the current user's speech rate is rapid. Through model training, it can intelligently generate a speech rate variation coefficient based on two key features: a reference value for volume fluctuation amplitude and a reference value for speech duration irregularity. This coefficient represents the degree of speech rate variation, and the system can use it to evaluate the user's speech rate in real time. If the speech rate variation coefficient is high, the model will determine that the speech rate is rapid and provide feedback for subsequent speech recognition adjustments. Through this process, the pre-trained machine learning model can significantly improve the system's accuracy in real-time speech recognition, especially when faced with fast or rapid speech, helping the system to respond more accurately and reduce the possibility of misrecognition.

[0100] The deep learning model is not limited here and can achieve the volume fluctuation amplitude reference value V index and speech duration irregularity reference value R duration Perform comprehensive analysis to generate the speech rate variation coefficient S peech In order to realize the technical solution of the present invention, the present invention provides a specific implementation method;

[0101] Speech rate variation coefficient S peech The generation formula is as follows:

[0102]

[0103] Where f1 and f2 are the reference values ​​of volume fluctuation amplitude V index and speech duration irregularity reference value R duration The preset proportional coefficient, and both f1 and f2 are greater than.

[0104] The preset proportional coefficients f1 and f2 are used to adjust the volume fluctuation amplitude reference value V index and speech duration irregularity reference value R duration Speech rate variation coefficient S peech The weight of the contribution. Specifically, f1 and f2 determine the relative importance of these two features in the final calculation. Their size determines the extent to which volume fluctuations and irregularities in speech duration affect the speech rate variation coefficient. Usually, these two coefficients need to be adjusted based on actual speech data so that the model can make more accurate assessments based on different speech characteristics. In other words, the preset proportional coefficients are set during the model training process and can be adjusted according to different tasks or application scenarios to optimize the accuracy and flexibility of speech rate assessment.

[0105] It can be seen from the speech rate variation coefficient that the larger the volume fluctuation amplitude reference value generated after analyzing the change of speech volume in unit time under the detection window, and the larger the speech duration irregularity reference value generated after analyzing the unevenness of the duration between words under the detection window, the larger the speech rate variation coefficient generated when the current user's speech rate is intelligently evaluated by the pre-trained deep learning model, indicating that the current user's speech speed is too fast, otherwise, it indicates that the current user's speech speed is stable.

[0106] The speech rate variation coefficient generated by the pre-trained deep learning model when intelligently evaluating the current user's speech rate is compared with the pre-set speech rate variation coefficient reference threshold to classify the current user's speech rate. The specific steps are as follows:

[0107] If the speech rate variation coefficient is greater than the preset speech rate variation coefficient reference threshold, the current user speech rate is classified as rapid speech; if the speech rate variation coefficient is less than or equal to the preset speech rate variation coefficient reference threshold, the current user speech rate is classified as slow speech.

[0108] When the deep learning model assesses that the user is speaking rapidly, it makes adaptive adjustments based on the user's voice input frequency and speed to reduce sensitivity to rapid speech, making it more tolerant of rapid speech input and reducing the probability of misrecognition. In addition, the processing time of the voice input is extended according to the rapidity of the speech, providing more processing space for the speech recognition process, thus ensuring that even rapid speech can be accurately parsed.

[0109] When the deep learning model assesses that the user is speaking rapidly, adaptive adjustments based on the frequency and speed of the user's voice input are crucial. This step is primarily intended to optimize the speech recognition system's responsiveness and accuracy when faced with rapid speech input. In the case of rapid speech, the speech recognition system may be unable to capture all the details of the rapid input in a timely manner, resulting in misidentification or missed recognition, impacting the system's effectiveness and user experience. Through adaptive adjustments, the system can reduce its oversensitivity to rapid speech, allowing it to handle rapid speech input with greater tolerance. This tolerance not only helps reduce the system's probability of misjudgment when faced with rapid speech, but also avoids unnecessary errors caused by excessive screening and strict requirements for certain details of speech input.

[0110] Another key measure is to extend the processing time of speech input based on the rapidity of speech. When speech is input too quickly, the system may not have enough time to accurately recognize each word or speech feature. By extending the processing time, the system has more processing space, allowing each part of the speech signal sufficient time to be accurately decoded, thus avoiding speech omissions and misunderstandings caused by time constraints.

[0111] This overall adjustment strategy, by incorporating the rapidity of speech, enables more efficient and accurate speech recognition. It improves the system's adaptability to rapid speech, ensuring high recognition accuracy across a wide range of speech speeds and reducing recognition errors caused by excessively fast speech. This not only enhances the system's fault tolerance but also improves the user experience, allowing users to receive accurate feedback and responses even when speaking rapidly.

[0112] When the deep learning model assesses that the user is speaking rapidly, adaptive adjustments are made based on the user's voice input frequency and speaking speed to reduce sensitivity to rapid speech. Furthermore, the specific steps for extending the processing time of voice input based on the rapidity of the speech are as follows:

[0113] When the deep learning model determines that the user is speaking rapidly, it will reduce its sensitivity to voice input and reduce misrecognition caused by rapid speech. The calculation expression is as follows:

[0114]

[0115] Where, α new is the adjusted sensitivity, α base is the default sensitivity, S peech is the speech rate variation coefficient, S ref is the reference threshold of speech rate variation coefficient, S maxis the maximum speech rate variation coefficient, ω is a nonlinear adjustment parameter that controls the degree of the inverse relationship between the speech rate variation coefficient and sensitivity. When ω>1, the sensitivity decreases more significantly when the speech rate is rapid. δ is the amplification factor of the sensitivity adjustment, which can provide stronger regulation for rapid speech rate.

[0116] Adjusted sensitivity α new It decreases significantly as S increases, ensuring that the system does not overfit to small changes in the voice input when the speaking speed is fast, making the system more tolerant to fast voice input and reducing the probability of misrecognition.

[0117] When a rapid speech rate is detected, the processing time of the speech input is dynamically extended according to the rapidity of the speech rate. The extended processing time is proportional to the speech rate change coefficient S. peech There is a nonlinear relationship between them and the reference threshold S of the speech rate variation coefficient. ref The calculation formula is as follows:

[0118]

[0119] Where, T ext is the extended processing time, T base is the default speech processing time, κ and ∈ are nonlinear adjustment coefficients. κ controls the extension of processing time when the speech rate is rapid, and ∈ controls the nonlinear relationship between speech rate changes and processing time extension. κ and ∈ are both greater than 1. λ is a smoothing coefficient used to adjust the smooth transition of speech rate rapidity to processing time extension.

[0120] This step extends the processing time so that the system can better capture fast-paced speech content and avoid missed or misidentified speech due to insufficient time.

[0121] By adjusting the sensitivity of the voice input and extending the processing time, the voice recognition process is automatically optimized when rapid speech is detected. peech and the speech rate variation coefficient reference threshold S ref Dynamic adjustments make the system more tolerant to fast voice input, while also ensuring longer voice processing time to improve voice recognition accuracy. The overall effect of the comprehensive adjustments is quantified by the following formula:

[0122]

[0123] Where RecognitionAccuracy is the accuracy of speech recognition, which is based on the adjusted sensitivity α new , extended processing time T ext and the speech rate variation coefficient S peechOptimization is carried out to maximize the accuracy of speech recognition and ensure efficient and accurate recognition results at different speaking speeds. is the index of the impact of rapid speaking on recognition accuracy, which controls the degree of influence of rapid speaking on recognition accuracy. σ is the amplification coefficient of the impact of speech rate changes on recognition accuracy, which is used to control the degree of influence of speech rate changes on recognition accuracy. ρ is the smoothing coefficient, which helps the system smoothly adjust the recognition process when speaking at a rapid speed and avoid overreaction caused by extreme adjustments.

[0124] Through comprehensive adjustments, the system can flexibly respond to changes in user speaking speed, ensuring stable and accurate recognition effects even when the speaking speed is fast.

[0125] The present invention effectively solves the problem of speech recognition errors caused by rapid speech speed by capturing user voice in real time and dynamically analyzing changes in speech speed. It is particularly suitable for scenarios involving professional terminology and complex guidance content, such as agricultural information processing. The method uses a deep learning model to intelligently evaluate speech speed, and makes adaptive adjustments based on the user's voice input frequency and the rapidity of the speech speed, reducing sensitivity to rapid speech while extending the processing time of voice input. Through these adjustments, the present invention ensures that the system can accurately understand user needs when faced with fast speech, reduces the probability of misrecognition, and improves recognition accuracy. Ultimately, farmers can obtain more accurate agricultural production recommendations, avoid decision-making errors caused by misleading information, reduce resource waste and agricultural production losses, and thus improve agricultural production efficiency and service quality.

[0126] The present invention provides Figure 2 The agricultural information management system based on virtual digital humans and AI shown in the figure includes a voice data acquisition module, a voice data preprocessing module, a speech rate feature extraction and dynamic analysis module, a deep learning evaluation module, and an adaptive adjustment and processing time extension module:

[0127] The voice data acquisition module first captures user voice input data in real time, collects user voice parameters, and provides raw data support for subsequent analysis and processing;

[0128] The voice data preprocessing module aggregates the acquired voice data parameters into an analysis set and preprocesses the collected voice data;

[0129] The speech rate feature extraction and dynamic analysis module extracts key features reflecting speech rate changes from preprocessed speech data and dynamically analyzes the extracted key features within the detection window to quantify the user's speech rate changes.

[0130] The deep learning evaluation module inputs the quantified features into a pre-trained deep learning model and uses the pre-trained deep learning model to intelligently evaluate the current user's speaking speed;

[0131] The adaptive adjustment and processing time extension module makes adaptive adjustments based on the user's voice input frequency and speed to reduce sensitivity to rapid speech when the deep learning model assesses that the user is speaking rapidly, making it more tolerant of rapid speech input and reducing the probability of misrecognition. In addition, the processing time of voice input is extended according to the rapidity of the speech rate, providing more processing space for the speech recognition process, thereby ensuring that even rapid speech can be accurately parsed.

[0132] An agricultural information processing method based on virtual digital humans and AI provided in an embodiment of the present invention is implemented through the above-mentioned agricultural information processing system based on virtual digital humans and AI. The specific methods and processes of an agricultural information processing system based on virtual digital humans and AI are detailed in the embodiment of the above-mentioned agricultural information processing method based on virtual digital humans and AI, which will not be repeated here.

[0133] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0134] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0135] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0136] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0137] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0138] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0139] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0140] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0141] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0142] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. An agricultural information management method based on virtual digital humans and AI, characterized in that: The following steps are involved: First, it captures user voice input data in real time, collects user voice parameters, and provides raw data support for subsequent analysis and processing; Aggregate the acquired voice data parameters into an analysis set and pre-process the collected voice data; Extract key features reflecting speech rate changes from preprocessed speech data and dynamically analyze the extracted key features within the detection window to quantify the user's speech rate changes. The quantified features are input into a pre-trained deep learning model, and the pre-trained deep learning model is used to intelligently evaluate the current user's speaking speed. When the deep learning model assesses that the user is speaking rapidly, it makes adaptive adjustments based on the user's voice input frequency and speaking speed to reduce sensitivity to rapid speech, making it more tolerant of rapid speech input and reducing the probability of misrecognition. In addition, the processing time of the voice input is extended according to the rapidity of the speaking speed, providing more processing space for the speech recognition process, thereby ensuring that even rapid speech can be accurately parsed.

2. The agricultural information management method based on virtual digital human and AI according to claim 1 is characterized in that: The specific steps for aggregating the acquired voice data parameters into an analysis set and performing preprocessing include: First, collect user voice data and extract multiple key parameters; Then, the parameters are aggregated into an analysis set in chronological order to form structured data for subsequent analysis; Next, the speech data is preprocessed, including noise removal, signal normalization, and removal of useless data to ensure the clarity and consistency of the audio data; During pre-processing, automatic speech segmentation is performed to effectively handle long speech inputs. The preprocessed data will provide high-quality basic data for subsequent feature extraction, speech rate analysis, and deep learning model input.

3. The agricultural information management method based on virtual digital human and AI according to claim 1 is characterized in that: Key features reflecting changes in speech rate are extracted from the preprocessed speech data. The extracted features include the change in speech volume per unit time and the unevenness of the duration between words. Within the detection window, the change in speech volume per unit time and the unevenness of the duration between words are analyzed and processed to generate reference values ​​for volume fluctuation amplitude and speech duration irregularity, respectively. The volume fluctuation amplitude reference value and speech duration irregularity reference value are used to quantify the degree of speech urgency and rhythm instability.

4. The agricultural information management method based on virtual digital human and AI according to claim 3 is characterized in that: The specific steps for analyzing the change in speech volume per unit time within the detection window to generate a reference value for the volume fluctuation amplitude are as follows: First, perform frame processing on the speech signal, divide the speech signal into multiple small time windows, and calculate the volume change in each window. The calculation expression is as follows: Where x i，j is the audio amplitude of the jth sampling point in the i-th frame, n is the number of sampling points in the current frame, V i is the instantaneous fluctuation amplitude of volume change; Get the instantaneous fluctuation amplitude V of the volume change i Finally, the volume fluctuation amplitude is quantified and a volume fluctuation amplitude reference value is generated. The quantization formula is as follows: Where V index is the volume fluctuation amplitude reference value, T is the total duration of the speech signal, N is the total number of signal frames, max(V) is the maximum value of the volume fluctuation amplitude, and α is the weighting coefficient used to control the degree of influence of the volume fluctuation amplitude on the final volume fluctuation reference value.

5. The agricultural information management method based on virtual digital human and AI according to claim 3 is characterized in that: The specific steps for analyzing the unevenness of the duration between words in the detection window to generate a reference value for speech duration irregularity are as follows: In this stage, we first extract the time interval between words in the speech signal. By framing and phoneme analysis of the speech signal, we accurately detect the start and end time of each word, and then calculate the duration difference between words. For each pair of words, we calculate the duration difference, that is, the time interval between words. The calculation expression is as follows: ΔT q =|T end，q -T start，q+1 | Where, ΔT q is the time difference between the qth word pair, that is, the duration time interval between two adjacent words, T end，q is the ending time of the qth word, T start，q+1 is the starting time of the q+1th word; After calculating the time difference ΔT between word pairs q Finally, we further quantify the fluctuation range of the difference and adopt the weighted reference value model to quantify its volatility through the nonlinear function of the interval time difference between word pairs. The formula is as follows: Where W q is the reference value of the fluctuation amplitude of the qth word pair, max(ΔT) is the maximum time difference between all word pairs in the entire speech signal, count q is the frequency of the qth word pair in the speech signal, θ and β are adjustment coefficients, θ is used to control the intensity of the influence of the time difference on the reference value of the fluctuation amplitude, and β is used to control the influence of the word pair frequency on the reference value of the fluctuation amplitude; Finally, by summarizing the fluctuation amplitudes of all word pairs, a reference value of speech duration irregularity is generated. The generation formula is as follows: Where R duration is the reference value of speech duration irregularity, M is the total number of word pairs in the speech signal, exp(-γ(ΔT q -ρ)) is the reference value attenuation function, γ is the reference value attenuation coefficient, which controls the steepness of the reference value function, and ρ is the threshold of the time difference.

6. The agricultural information management method based on virtual digital human and AI according to claim 3 is characterized in that: The analyzed volume fluctuation amplitude reference value and speech duration irregularity reference value are input into a pre-trained deep learning model, and a speech rate variation coefficient is generated by the deep learning model. The current user's speech rate is intelligently evaluated by the speech rate variation coefficient.

7. The agricultural information management method based on virtual digital human and AI according to claim 6 is characterized in that: The speech rate variation coefficient generated by the pre-trained deep learning model when intelligently evaluating the current user's speech rate is compared with the pre-set speech rate variation coefficient reference threshold to classify the current user's speech rate. The specific steps are as follows: If the speech rate variation coefficient is greater than the preset speech rate variation coefficient reference threshold, the current user speech rate is classified as rapid speech; if the speech rate variation coefficient is less than or equal to the preset speech rate variation coefficient reference threshold, the current user speech rate is classified as slow speech.

8. The agricultural information management method based on virtual digital human and AI according to claim 7 is characterized in that: When the deep learning model assesses that the user is speaking rapidly, adaptive adjustments are made based on the user's voice input frequency and speaking speed to reduce sensitivity to rapid speech. Furthermore, the specific steps for extending the processing time of voice input based on the rapidity of the speech are as follows: When the deep learning model determines that the user is speaking rapidly, it will reduce its sensitivity to voice input and reduce misrecognition caused by rapid speech. The calculation expression is as follows: Where, α new is the adjusted sensitivity, α base is the default sensitivity, S peech is the speech rate variation coefficient, S ref is the reference threshold of speech rate variation coefficient, S max is the maximum speech rate variation coefficient, ω is a nonlinear adjustment parameter that controls the degree of the inverse relationship between the speech rate variation coefficient and sensitivity, and δ is the amplification factor of the sensitivity adjustment, which can provide a stronger adjustment for rapid speech rate; When a rapid speech rate is detected, the processing time of the speech input is dynamically extended according to the rapidity of the speech rate. The extended processing time is proportional to the speech rate change coefficient S. peech There is a nonlinear relationship between them and the reference threshold value S of the speech rate variation coefficient. ref The calculation formula is as follows: Where, T ext is the extended processing time, T base is the default speech processing time, κ and ∈ are nonlinear adjustment coefficients, κ controls the extension of processing time when the speech rate is fast, ∈ controls the nonlinear relationship between speech rate change and processing time extension, κ and ∈ are both greater than 1, λ is the smoothing coefficient; By adjusting the sensitivity of the voice input and extending the processing time, the voice recognition process is automatically optimized when rapid speech is detected. peech and the speech rate variation coefficient reference threshold S ref Dynamic adjustments make the system more tolerant to fast voice input, while also ensuring longer voice processing time to improve voice recognition accuracy. The overall effect of the comprehensive adjustments is quantified by the following formula: Where RecognitionAccuracy is the accuracy of speech recognition, is the impact index of rapid speaking rate on recognition accuracy, σ is the amplification coefficient of the impact of speech rate change on recognition accuracy, and ρ is the smoothing coefficient.

9. An agricultural information management system based on virtual digital humans and AI, used to implement the agricultural information management method based on virtual digital humans and AI as described in any one of claims 1 to 8, characterized in that: It includes voice data acquisition module, voice data preprocessing module, speech rate feature extraction and dynamic analysis module, deep learning evaluation module and adaptive adjustment and processing time extension module: The voice data acquisition module first captures user voice input data in real time, collects user voice parameters, and provides raw data support for subsequent analysis and processing; The voice data preprocessing module aggregates the acquired voice data parameters into an analysis set and preprocesses the collected voice data; The speech rate feature extraction and dynamic analysis module extracts key features reflecting speech rate changes from preprocessed speech data and dynamically analyzes the extracted key features within the detection window to quantify the user's speech rate changes. The deep learning evaluation module inputs the quantified features into a pre-trained deep learning model and uses the pre-trained deep learning model to intelligently evaluate the current user's speaking speed; The adaptive adjustment and processing time extension module makes adaptive adjustments based on the user's voice input frequency and speed to reduce sensitivity to rapid speech when the deep learning model assesses that the user is speaking rapidly, making it more tolerant of rapid speech input and reducing the probability of misrecognition. In addition, the processing time of voice input is extended according to the rapidity of the speech rate, providing more processing space for the speech recognition process, thereby ensuring that even rapid speech can be accurately parsed.

Images