Voice interaction graphene AI intelligent tea table control method

Abstract

The invention discloses a voice interaction graphene AI intelligent tea table control method, and relates to the technical field of home voice interaction, and the method comprises the control steps: S1, monitoring a surrounding user instruction based on a built-in microphone of a tea table, and after the tea table receives the voice instruction, synchronously carrying out the preliminary noise reduction processing, and retaining the voice segment of the instruction; s2, training a data set containing a noise scene, enabling a built-in model to learn features for distinguishing voice and noise, and filtering non-voice signals in real time; and S3, for dialects and accents in different regions, collecting a scale data set, training a dialect recognition model, and inputting the obtained voice signal into the dialect recognition model to recognize and judge the current dialect and accents. Through preliminary noise reduction and microphone array layout, the voice acquisition quality is improved, and instruction omission is reduced; secondly, by training a multi-noise scene data set, the adaptability of the model to a complex environment is enhanced, and reliable recognition is guaranteed; and aiming at dialects and accent optimization, the product universality is improved.

Description

A voice-interactive graphene AI smart coffee table control method Technical Field

[0001] This invention relates to the field of home voice interaction technology, specifically to a voice-interactive graphene AI smart coffee table control method. Background Technology

[0002] With the rapid development of smart home technology, various smart terminal devices are gradually being integrated into people's daily lives. Traditional coffee tables only have the single function of supporting items, while modern graphene coffee tables are a new type of home furnishing product that integrates the characteristics of graphene materials with intelligent functions. They retain the supporting function of traditional coffee tables as the core furniture of the living room, and expand more practical value through the unique advantages of graphene materials and the integration of smart technology. They have gradually integrated touch operation, information display, and environmental monitoring functions, but there are still many limitations in human-computer interaction. The current mainstream smart coffee table interaction methods mostly rely on physical buttons or touch screens, and users need to be in close contact with the device to complete the operation, which is extremely inconvenient when their hands are occupied or relaxed. Voice interaction, as a natural and convenient human-computer interaction method, has been widely used in smartphones and smart speaker devices. However, its application in specific scenarios such as smart coffee tables still faces many technical challenges. Environmental noise is one of the important factors affecting the accuracy of voice recognition. For example, the sound of a TV playing or the conversations of people around can interfere with voice commands, resulting in inaccurate control commands received by the smart coffee table and the inability to achieve effective voice interaction. Furthermore, the user's pronunciation, dialect, or accent may also prevent the voice recognition system from accurately recognizing all commands. When people from different regions use voice control, the recognition rate varies greatly, leading to recognition errors and the inability to achieve effective interaction. Therefore, there is an urgent need to propose an AI smart coffee table control method with voice interaction. Summary of the Invention

[0003] To address the aforementioned technical issues, a voice-interactive graphene AI smart coffee table control method is provided, which solves the problems described above.

[0004] To achieve the above objectives, the technical solution adopted by this invention is as follows: a voice-interactive graphene AI smart coffee table control method, comprising the following control steps: S1, using the coffee table's built-in microphone to listen for user commands from the surrounding area; upon receiving a voice command, the coffee table simultaneously performs preliminary noise reduction processing, preserving the voice segment of the command; S2, training a dataset containing noisy scenes, allowing the built-in model to learn the characteristics that distinguish between speech and noise, and filtering non-speech signals in real time; S3, collecting large-scale datasets for different regional dialects and accents, training a dialect recognition model, inputting the acquired voice signal into the dialect recognition model to identify and judge the current dialect and accent, and performing text conversion; S4, parsing the converted text, analyzing the intent and command information; if the command is ambiguous, actively asking the user to clarify their needs; if there are multiple intent commands, breaking them down into sub-commands for sequential execution; and transmitting the command to the corresponding functional module to perform the interactive operation.

[0005] Preferably, in step S1, the smart coffee table has a built-in microphone array distributed along the edge and bottom of the coffee table; the noise reduction process uses the hardware's built-in low-pass filter to filter out high-frequency electromagnetic interference above 16kHz and low-frequency vibration noise below 20Hz, retaining signals in the 200Hz-8kHz frequency band.

[0006] Preferably, the specific steps in step S2 are as follows: During the training phase, a noisy speech dataset is first constructed. This is achieved by collecting clean speech and various noise sample data, mixing them with a signal-to-noise ratio ranging from -5dB to 20dB, and dividing the dataset into training, validation, and test sets in a 7:2:1 ratio, while also labeling the speech region timestamps. The audio is then converted into frequency domain features to obtain a two-dimensional feature matrix, which visually represents the spectral differences between speech and noise. A CNN is selected as the model architecture, and the model is trained to minimize the loss, maximizing the model's accuracy in predicting speech frames as 1 and noise frames as 0. The Adam optimizer is used with early stopping and data augmentation to avoid overfitting. The model output is evaluated using speech recognition rate and noise filtering rate as evaluation metrics. The trained model is then deployed to the coffee table system for real-time filtering. The audio stream is collected and cut into segments of fixed frame length. Features from the two-dimensional feature matrix are extracted and input into the model to obtain frame-by-frame speech probabilities. After smoothing via a sliding window, a threshold is used to determine speech and noise regions, and the denoised speech is output.

[0007] Preferably, the duration of speech in the sample data is 1-10 seconds. The mixed sample is generated by superimposing speech and noise in the time domain using an audio processing library. The superposition formula is: Mixed signal = Clean speech signal * Superposition coefficient 1 + Noise signal * Superposition coefficient 2. The superposition coefficient is obtained by determining the target signal-to-noise ratio according to the scene, calculating the sum of the energy of clean speech and noise, fixing the speech coefficient, and deriving the noise coefficient. The clean speech and noise are quantized by converting the continuous analog signal of the collected sample data into a discrete digital signal to obtain the quantized value, which is the clean speech signal and the noise signal.

[0008] Preferably, the steps for obtaining the two-dimensional feature matrix are as follows: the continuous audio in the mixed samples is divided into short frames of 20-30ms with a 10ms overlap between frames; a Hanning window is applied to each frame signal, and a fast Fourier transform is performed on each windowed frame to convert the time-domain signal into the frequency domain, obtaining the amplitude of each frequency component and forming a spectrum; a Mel filter bank is used to simulate the nonlinear perception of frequency by the human ear, mapping the spectrum to the Mel frequency scale and preserving key frequency band information of the speech; the logarithm of the Mel spectrum is taken to compress the dynamic range and enhance the recognition of low-energy features, resulting in a two-dimensional feature matrix in which speech exhibits regular spectral stripes, while noise appears to be randomly distributed.

[0009] Preferably, the specific steps in step S3 are as follows: Identify the included dialect types; collect speech data using recording equipment; unify the collected speech data into WAV and MP3 formats; set the sampling rate to 16kHz; remove data with labeling errors and poor audio quality; extract speech features by processing the data through framing, windowing, Fourier transform, Mel filtering, and discrete cosine transform; divide the processed data proportionally for model training, validation, and evaluation; optimize the model, with the optimization steps being the same as the feature optimization steps in step S2 where the built-in model learns to distinguish speech from noise; deploy the optimized model on a cloud server; after acquiring the speech signal, preprocess it and input it into the dialect recognition model; the model outputs the recognition result, determines the dialect and accent of the current speech, and converts the speech into corresponding text content.

[0010] Preferably, the cloud server is an 8-core, 16GB memory cloud server. The optimized model is stored in ONNX format, and the model initialization time is less than 3 seconds. Preprocessing includes format verification and conversion. After receiving the audio, the format is first verified. If it is MP3 format, it is converted to WAV format in real time using FFmpeg. CRC verification is enabled during the conversion process to ensure data integrity. After inputting into the dialect recognition model, the long-term dependencies of the feature sequence are captured by a 6-layer self-attention mechanism. A 15-dimensional logits are output through a fully connected layer. The Softmax function is applied to convert the logits into a probability distribution, satisfying that the sum of all probabilities is 1. The top 3 candidate categories with the highest probabilities and their probability values ​​are extracted. If the highest probability is ≥0.6, the corresponding category is directly used as the recognition result. If the highest probability is <0.6, a second inference is triggered. The 500ms segment with the highest energy in the audio is extracted again to re-extract features. The input is fed into the model to calculate the probability again. The highest average probability value of the two inferences is taken as the final category.

[0011] Preferably, the steps in S4 for analyzing sentence intent and instruction information are as follows: Clean redundant symbols, retain core semantic components, unify encoding, and map dialect words to general semantics; perform word segmentation, using a dialect adaptation tool to split the text and handle adjoining words; label each segmented result with part-of-speech tags, focusing on instruction-related parts of speech; construct an intent classification system, pre-setting common intent categories to cover home control, information query, and social interaction scenarios, with basic intents including: turn on lights, turn off lights, adjust temperature, query weather, and play music; input the segmented text sequence, output the intent probability distribution, and resolve ambiguity based on context; extract key information from the identified instruction intents, including the operation object, operation action, attribute parameters, and time; for missing elements, complete the missing requirements through proactive inquiry; for compound instructions, progressively split and execute interactive operations step by step, and finally output the structured results for intelligent control.

[0012] Preferably, redundant symbols include modal particles and repetition symbols. The word segmentation process is as follows: For the standardized dialect text, a word segmentation tool is used for processing. The word segmentation tool, combined with the collected dialect samples, uses a bidirectional maximum matching method to split the text, starting from the left and right sides of the text respectively. It prioritizes matching the longest valid words in the dictionary, and then compares the splitting results in the two directions, selecting the sequence with fewer words as the initial word segmentation result. For adjoining words not included in the dictionary, the basic morphemes contained in the adjoining words are identified, and the adjoining words are split into two or more words that can independently express part of the semantics. Each word obtained by word segmentation is labeled with its corresponding part of speech. The labeling is based on the conventional grammatical function of the word in the dialect, and the dialect word class library in the big data is referenced during the labeling. For core words related to instructions, special marking is performed to highlight their functions.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: First, by implementing preliminary noise reduction and microphone array layout, this invention improves the quality of voice acquisition and reduces command omissions. Second, by training on multi-noise scene datasets, it enhances the model's adaptability to complex environments, reduces false triggers, and ensures reliable recognition. Third, it optimizes for dialects and accents, expanding the applicable population and improving product universality. Finally, the text parsing and intent analysis mechanism can handle ambiguous and multi-intent commands, ensuring accurate execution and precise transmission to the corresponding modules, improving the interactive experience. Overall, it is user-centric, achieving natural and intelligent interaction, making operation more convenient, fitting home scenarios, and bringing an efficient and user-friendly experience. Attached Figure Description

[0014] Figure 1 is a flowchart of the control steps of the present invention. Detailed Implementation

[0015] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0016] Referring to Figure 1, a voice-interactive graphene AI smart coffee table control method includes the following steps: S1. The coffee table uses its built-in microphone to listen for user commands from the surrounding area. After receiving a voice command, the coffee table performs preliminary noise reduction processing to retain the voice segment of the command. S2. A dataset containing noisy scenes is trained, allowing the built-in model to learn the characteristics that distinguish between speech and noise, and to filter non-speech signals in real time. S3. For dialects and accents from different regions, a large-scale dataset is collected to train a dialect recognition model. The acquired speech signal is input into the dialect recognition model to identify and judge the current dialect and accent, and then converts it into text. S4. The converted text is parsed, and the intent and command information of the statement are analyzed. If the command is ambiguous, the user is actively asked to clarify the needs. If there are multiple intent commands, they are broken down into sub-commands and executed sequentially. The commands are then passed to the corresponding functional modules to perform interactive operations.

[0017] First, by implementing initial noise reduction and microphone array layout, the quality of voice acquisition is improved, and command omissions are reduced. Second, by training on datasets of multiple noisy scenarios, the model's adaptability to complex environments is enhanced, false triggers are reduced, and recognition reliability is ensured. Third, optimizations for dialects and accents expand the applicable population and improve product universality. Finally, the text parsing and intent analysis mechanism can handle ambiguous and multi-intent commands, ensuring accurate execution and precise transmission to the corresponding modules, thus improving the interactive experience. Overall, the system is user-centric, achieving natural and intelligent interaction, making operation more convenient, fitting home scenarios, and bringing an efficient and user-friendly experience.

[0018] In step S1, the smart coffee table has a built-in microphone array distributed along its edges and bottom. The noise reduction process uses the hardware's built-in low-pass filter to filter out high-frequency electromagnetic interference above 16kHz and low-frequency vibration noise below 20Hz, retaining signals in the 200Hz-8kHz frequency band.

[0019] In step S1, the smart coffee table listens to user commands through a built-in microphone array. These microphones are distributed along the edges and bottom of the coffee table. This layout can capture voice information from different directions more comprehensively. The initial noise reduction processing, combined with hardware low-pass filtering, accurately filters high-frequency electromagnetic interference above 16kHz and low-frequency vibration sounds below 20Hz, retaining only the 200Hz-8kHz frequency band where human speech is mainly concentrated, providing a cleaner voice foundation for subsequent processing.

[0020] The specific steps in step S2 are as follows: During the training phase, a noisy speech dataset is first constructed. This is achieved by collecting clean speech and various noise samples, mixing them with a signal-to-noise ratio ranging from -5dB to 20dB, and dividing the dataset into training, validation, and test sets in a 7:2:1 ratio. Speech region timestamps are then labeled. The audio is converted to frequency domain features to obtain a two-dimensional feature matrix, which visually represents the spectral differences between speech and noise. A CNN is selected as the model architecture, and the model is trained to minimize loss, maximizing the accuracy of predicting 1 for speech frames and 0 for noise frames. The Adam optimizer is used with early stopping and data augmentation to avoid overfitting. The model output is evaluated using speech recognition rate and noise filtering rate as evaluation metrics. The trained model is then deployed to a dedicated system for real-time filtering. The audio stream is collected and segmented into fixed frame lengths, features are extracted from the two-dimensional feature matrix, and input into the model to obtain frame-by-frame speech probabilities. After smoothing via a sliding window, a threshold is used to determine speech and noise regions, and the denoised speech is output.

[0021] In step S2, constructing a noisy speech dataset is crucial for model training. Speech samples range from 1 to 10 seconds in length, covering common lengths of everyday commands. Samples are generated by mixing signals with noise ratios ranging from -5dB to 20dB, encompassing various environments from noisy to quiet. In the formula for calculating the mixed signal, the superposition coefficient is determined based on the target signal-to-noise ratio of the scene. By calculating the energy of speech and noise, and fixing the speech coefficient, the noise coefficient is derived, allowing for precise control of the mixed signal-to-noise ratio. Quantization converts continuous analog audio signals into discrete digital signals that can be processed by a computer, facilitating subsequent feature extraction and model training. The process of obtaining the two-dimensional feature matrix involves converting continuous audio signals into discrete digital signals. Frequency segmentation is divided into 20-30ms short frames with a 10ms overlap. This framing method ensures the independence of each frame while preserving the continuity between frames through overlap. The application of the Hanning window reduces spectral leakage, making the frequency domain information after Fourier transform more accurate. The Mel filter bank simulates the nonlinear perception characteristics of the human ear on frequency, highlighting key speech information. The logarithm of the Mel spectrum is then taken to further compress the dynamic range, making low-energy speech features easier to identify. At this point, the regular spectral stripes of speech in the two-dimensional feature matrix contrast sharply with the chaotic distribution of noise, which is beneficial for model differentiation.

[0022] The duration of speech in the sample data ranges from 1 to 10 seconds. The mixed samples are generated by superimposing speech and noise in the time domain using an audio processing library. The superposition formula is: Mixed signal = Clean speech signal * Superposition coefficient 1 + Noise signal * Superposition coefficient 2. The superposition coefficient is obtained by determining the target signal-to-noise ratio based on the scene, calculating the sum of the energy of the clean speech and noise, fixing the speech coefficient, and deriving the noise coefficient. The clean speech and noise are quantized by converting the continuous analog signal of the collected sample data into a discrete digital signal to obtain the quantized value, which is the clean speech signal and the noise signal.

[0023] The steps for obtaining the two-dimensional feature matrix are as follows: The continuous audio in the mixed samples is segmented into short frames of 20-30ms, with a 10ms overlap between frames; a Hanning window is applied to each frame signal, and a Fast Fourier Transform is performed on each windowed frame to convert the time-domain signal into the frequency domain, obtaining the amplitude of each frequency component and forming a spectrum; a Mel filter bank is used to simulate the nonlinear perception of frequency by the human ear, mapping the spectrum to the Mel frequency scale and preserving key frequency band information of the speech; the logarithm of the Mel spectrum is taken to compress the dynamic range and enhance the recognition of low-energy features, resulting in a two-dimensional feature matrix. In the matrix, speech exhibits regular spectral stripes, while noise appears randomly distributed.

[0024] The training and real-time deployment inference process of the model is as follows: Clean speech and noise are mixed, and diverse samples are generated according to different SNRs. Each frame is labeled with a tag: speech frame = 1, noise frame = 0. The audio is segmented into fixed frame lengths, such as 25ms / frame. A two-dimensional frequency domain feature matrix is ​​generated through temporal framing → windowing → FFT → Mel filtering → logarithmic transformation. A lightweight CNN network is designed to adapt to real-time requirements. The two-dimensional frequency domain feature matrix, such as a Mel spectrogram, with dimensions [time frame, Mel band, 1], is received and adapted to the CNN in a single channel. Multi-layer convolution (such as 3×3 convolution kernels) is used to extract local spectral features, capturing the differences in speech formants and the disordered distribution of noise. Dimensionality reduction is achieved through max pooling / average pooling, preserving key features while compressing computation. The convolutional features are mapped to binary classification, speech / noise, and the frame-by-frame prediction probability (0~1) is output. Binary cross-entropy loss is used to quantize the prediction probability and the true label (1 / 0). The goal is to minimize the loss based on the differences between audio frames and noise. The Adam optimizer is used to dynamically adjust the learning rate, and during training, audio speed (±10%), volume (±20%), and random noise are added to increase data diversity. The validation set loss is monitored; if it does not decrease for five consecutive rounds, training is stopped, and the optimal model is retained. The proportion of speech frames correctly predicted as 1 and the proportion of noise frames correctly predicted as 0 are measured. Audio streams are acquired via a microphone / audio interface and segmented into fixed frame lengths. The feature transformation process (temporal segmentation → windowing → FFT → Mel filtering → logarithmic transform) is repeated for each audio frame to generate a two-dimensional feature matrix, which is then used as model input. This feature matrix is ​​then input into the CNN model. Output frame-by-frame speech probabilities; filter the probability sequence using a sliding window of 3-5 frames to avoid misclassification in a single frame; set a probability threshold, and regions with a probability greater than the threshold are considered speech regions, otherwise they are considered noise regions; for regions identified as noise, suppress noise through spectral subtraction / energy attenuation; concatenate the processed audio frames into a continuous stream and output the denoised speech; during training, the shape of the feature matrix needs to be clearly defined to ensure that the CNN input dimension matches; during real-time inference, the dimensions of single-frame features need to be expanded to adapt to the model input; the window length needs to be adjusted according to the scenario; the smoothing method can be flexibly selected to balance response speed and misclassification rate; a fixed threshold may not be suitable for all scenarios, and an adaptive threshold can be designed to improve robustness.

[0025] Inference speed needs to be optimized through model quantization and hardware acceleration to ensure that the end-to-end latency (acquisition → processing → output) is less than 100ms, in order to meet the requirements of real-time interaction.

[0026] The specific steps in step S3 are as follows: First, identify the included dialect types. Second, collect speech data using recording equipment, unifying the collected speech data into WAV and MP3 formats, and setting the sampling rate to 16kHz. Third, remove data with labeling errors and poor audio quality, and extract speech features through framing, windowing, Fourier transform, Mel filtering, and discrete cosine transform. Fourth, divide the processed data proportionally for model training, validation, and evaluation. Fifth, optimize the model; the optimization steps are the same as those in step S2, where the built-in model learns to distinguish speech from noise features, and will not be elaborated further. Sixth, deploy the optimized model on a cloud server. After acquiring the speech signal, preprocess it and input it into the dialect recognition model. The model outputs the recognition result, determines the dialect and accent of the current speech, and converts the speech into corresponding text content.

[0027] This application clearly identifies dialect types and collects data specifically to avoid irrelevant data interference and ensure the targeted nature of model training; it unifies the audio format (WAV / MP3) and sampling rate (16kHz) to reduce the impact of format differences on model learning and ensure data consistency; it removes erroneous and low-quality data, which is equivalent to "purifying" the training data, reducing the misleading influence of noisy data on the model and allowing the model to focus more on effective features; it extracts speech features through professional methods such as frame segmentation and Fourier transform, which can accurately capture the key information of tone and rhythm unique to dialects, making it easier for the model to distinguish the differences between different dialects / accents; it divides the training, validation, and evaluation sets proportionally, which conforms to the scientific process of machine learning, making it easy to adjust parameters through the validation set and check the effect through the evaluation set, avoiding overfitting or underfitting of the model, and ensuring stable performance in real-world scenarios.

[0028] The cloud server is an 8-core, 16GB memory cloud server. The optimized model is stored in ONNX format, and the model initialization time is less than 3 seconds. Preprocessing includes format verification and conversion. After receiving the audio, the format is first verified. If it is in MP3 format, it is converted to WAV format in real time using FFmpeg. CRC check is enabled during the conversion process to ensure data integrity. After inputting into the dialect recognition model, a 6-layer self-attention mechanism is used to capture the long-term dependencies of the feature sequence. A 15-dimensional logits is output through a fully connected layer. The Softmax function is applied to convert the logits into a probability distribution, satisfying that the sum of all probabilities is 1. The top 3 candidate categories with the highest probabilities and their probability values ​​are extracted. If the highest probability is ≥0.6, the corresponding category is directly used as the recognition result. If the highest probability is <0.6, a second inference is triggered. The 500ms segment with the highest energy in the audio is extracted again to re-extract features. The input is fed into the model to calculate the probability again. The highest average probability value of the two inferences is taken as the final category.

[0029] In the S3 step of dialect recognition, the collected speech data is uniformly in WAV and MP3 formats with a sampling rate of 16kHz to ensure data format consistency. Format verification and conversion in preprocessing are crucial. CRC verification is enabled when converting MP3 to WAV to effectively ensure the integrity of the transcoded data. An 8-core, 16GB memory cloud server provides sufficient computing power for model operation. The model initialization time for ONNX format storage is less than 3 seconds, ensuring response speed. The 6-layer self-attention mechanism can effectively capture long-term dependencies in speech feature sequences, improving the accuracy of dialect recognition. When the highest probability is <0.6, the 500ms segment with the highest energy in the audio is extracted for secondary inference. This segment usually contains the clearest speech information. The final category is determined by the average probability of the two inferences, further improving the reliability of recognition.

[0030] The steps in S4 for analyzing sentence intent and instruction information are as follows: Clean redundant symbols, retain core semantic components, unify encoding, and map dialect words to universal semantics; perform word segmentation, using a dialect adaptation tool to split the text and handle adjoining words; label each segmented result with part-of-speech tags, focusing on instruction-related parts of speech; construct an intent classification system, pre-setting common intent categories to cover home control, information query, and social interaction scenarios, with basic intents including: turn on lights, turn off lights, adjust temperature, query weather, and play music; input the segmented text sequence, output the intent probability distribution, and resolve ambiguities based on context; extract key information from the identified instruction intents, including the target object, operation action, attribute parameters, and time; for missing elements, complete the missing requirements through proactive inquiry; for compound instructions, progressively break them down and execute the interactive operation step by step, finally outputting the structured results for intelligent control.

[0031] Redundant symbols include modal particles and repetition symbols. The word segmentation process is as follows: For the standardized dialect text, a word segmentation tool is used for processing. The word segmentation tool, combined with the collected dialect samples, uses a bidirectional maximum matching method to split the text, starting from the left and right sides of the text respectively. It prioritizes matching the longest valid words in the dictionary, and then compares the splitting results in the two directions, selecting the sequence with fewer words as the initial word segmentation result. For adjoining words not included in the dictionary, the basic morphemes contained in the adjoining words are identified, and the adjoining words are split into two or more words that can independently express part of the semantics. Each word obtained from word segmentation is labeled with its corresponding part of speech. The labeling is based on the conventional grammatical function of the word in the dialect, and the dialect word class library in the big data is referenced during the labeling. For core words related to instructions, special marking is performed to highlight their functions.

[0032] In the S4 step's analysis of sentence intent and instruction information, cleaning redundant symbols simplifies the text and highlights the core semantics. The word segmentation process uses a bidirectional maximum matching method, combined with a dialect sample lexicon, to simultaneously match the longest effective words from both sides, and then select the segmentation result with fewer words, which can more accurately segment dialect text. For unrecorded adjoining words, they are segmented by identifying basic morphemes, ensuring the integrity of word segmentation. Relying on the common categories preset by the classification system, it comprehensively covers home control, information query, and social interaction scenarios. When extracting key information, if there are missing elements, the system will actively ask for completion. For compound instructions, they are split into sub-instructions and executed sequentially to ensure the accuracy and orderliness of instruction execution, ultimately achieving precise and intelligent control of the smart coffee table.

[0033] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for controlling a graphene AI smart coffee table with voice interaction, characterized in that, The control steps are as follows: S1. Based on the built-in microphone of the coffee table, listen to the surrounding user commands. After receiving the voice command, the coffee table performs preliminary noise reduction processing and retains the voice segment of the command. S2. Train the dataset containing noisy scenes, so that the built-in model learns to distinguish the features of speech and noise and filters non-speech signals in real time. S3. Collect large-scale datasets for dialects and accents in different regions, train the dialect recognition model, input the acquired speech signal into the dialect recognition model to recognize and judge the current dialect and accent, and perform text conversion. S4. Parse the converted text, analyze the intent and instructions. If the instructions are ambiguous, actively ask the user to clarify their needs. If there are multiple intents, break them down into sub-instructions and execute them sequentially. Pass the instructions to the corresponding functional modules to perform interactive operations.

2. The voice-interactive graphene AI smart coffee table control method according to claim 1, characterized in that: In step S1, the smart coffee table has a built-in microphone array distributed along its edges and bottom. The noise reduction process uses the hardware's built-in low-pass filter to filter out high-frequency electromagnetic interference above 16kHz and low-frequency vibration noise below 20Hz, retaining signals in the 200Hz-8kHz frequency band.

3. The method for controlling a graphene AI smart coffee table with voice interaction according to claim 1, characterized in that, The specific steps in step S2 are as follows: During the training phase, a noisy speech dataset is first constructed. This is done by collecting clean speech and various noise samples, mixing them with a signal-to-noise ratio ranging from -5dB to 20dB, and dividing the dataset into training, validation, and test sets in a 7:2:1 ratio. Speech region timestamps are then labeled. The audio is converted into frequency domain features to obtain a two-dimensional feature matrix, which visually represents the spectral differences between speech and noise. A CNN is selected as the model architecture, and the model is trained to minimize the loss, maximizing the accuracy of predicting 1 for speech frames and 0 for noise frames. The Adam optimizer is used with early stopping and data augmentation to avoid overfitting. The speech recognition rate and noise filtering rate are used as evaluation metrics to assess the model's output. The trained model is deployed to the coffee table system for real-time filtering. The audio stream is collected and cut into segments with a fixed frame length. Features in the two-dimensional feature matrix are extracted and input into the model to obtain the frame-by-frame speech probability. After smoothing by a sliding window, a threshold is used to determine the speech and noise regions, and the denoised speech is output.

4. The voice-interactive graphene AI smart coffee table control method according to claim 3, characterized in that: The duration of speech in the sample data ranges from 1 to 10 seconds. The mixed samples are generated by superimposing speech and noise in the time domain using an audio processing library. The superposition formula is: Mixed signal = Clean speech signal * Superposition coefficient 1 + Noise signal * Superposition coefficient 2. The superposition coefficient is obtained by determining the target signal-to-noise ratio based on the scene, calculating the sum of the energy of the clean speech and noise, fixing the speech coefficient, and deriving the noise coefficient. The clean speech and noise are quantized by converting the continuous analog signal of the collected sample data into a discrete digital signal to obtain the quantized value, which is the clean speech signal and the noise signal.

5. The method for controlling a graphene AI smart coffee table with voice interaction according to claim 3, characterized in that, The steps for obtaining the two-dimensional feature matrix are as follows: The continuous audio in the mixed samples is segmented into short frames of 20-30ms, with a 10ms overlap between frames; a Hanning window is applied to each frame signal, and a Fast Fourier Transform is performed on each windowed frame to convert the time-domain signal into the frequency domain, obtaining the amplitude of each frequency component and forming a spectrum; a Mel filter bank is used to simulate the nonlinear perception of frequency by the human ear, mapping the spectrum to the Mel frequency scale and preserving key frequency band information of the speech; the logarithm of the Mel spectrum is taken to compress the dynamic range and enhance the recognition of low-energy features, resulting in a two-dimensional feature matrix. In the matrix, speech exhibits regular spectral stripes, while noise appears randomly distributed.

6. The method for controlling a graphene AI smart coffee table with voice interaction according to claim 1, characterized in that, The specific steps in step S3 are as follows: Identify the included dialect types; collect speech data using recording equipment; unify the collected speech data into WAV and MP3 formats; set the sampling rate to 16kHz; remove data with labeling errors and poor audio quality; extract speech features by processing the data through framing, windowing, Fourier transform, Mel filtering, and discrete cosine transform; divide the processed data proportionally for model training, validation, and evaluation; optimize the model, following the same optimization steps as those in step S2 where the built-in model learns to distinguish speech from noise features; deploy the optimized model on a cloud server; after acquiring the speech signal, preprocess it and input it into the dialect recognition model; the model outputs the recognition result, determines the dialect and accent of the current speech, and converts the speech into corresponding text content.

7. The voice-interactive graphene AI smart coffee table control method according to claim 6, characterized in that: The cloud server is an 8-core, 16GB memory cloud server. The optimized model is stored in ONNX format. The model initialization time is less than 3 seconds. Preprocessing includes format verification and conversion. After receiving the audio, the format is first verified. If it is MP3 format, it is converted to WAV format in real time using FFmpeg. CRC verification is enabled during the conversion process to ensure data integrity. After inputting into the dialect recognition model, the long-term dependencies of the feature sequence are captured through a 6-layer self-attention mechanism. The 15-dimensional logits are output through a fully connected layer. The Softmax function is applied to convert the logits into a probability distribution, satisfying that the sum of all probabilities is 1. The top 3 candidate categories with the highest probabilities and their probability values ​​are extracted. If the highest probability is ≥0.6, the corresponding category is directly used as the recognition result. If the highest probability is less than 0.6, a second inference is triggered. The 500ms segment with the highest energy in the audio is extracted again to re-extract features. The input is then fed into the model to calculate the probability again. The highest average probability value of the two inferences is taken as the final category.

8. The method for controlling a graphene AI smart coffee table with voice interaction according to claim 1, characterized in that, The steps in step S4 for analyzing the intent and instruction information of a statement are as follows: clean up redundant symbols, retain core semantic components, unify encoding, and map dialect words to general semantics; perform word segmentation, split the text using a dialect adaptation tool, and handle adjoining words; and label each word segmentation result with part-of-speech tags, with a focus on marking the part-of-speech tags related to the instruction. An intent classification system is constructed, pre-setting common intent categories to cover home control, information query, and social interaction scenarios. Basic intents include: turning lights on, turning lights off, adjusting temperature, checking the weather, and playing music. The input is a segmented text sequence, and the output is an intent probability distribution, which is combined with context to resolve ambiguity. Key information is extracted from the identified instruction intents, including the operation object, operation action, attribute parameters, and time. For missing elements, the missing requirements are completed through active inquiry. For compound instructions, the interactive operation is executed step by step after being broken down. Finally, the results are output in a structured manner for intelligent control.

9. A voice-interactive graphene AI smart coffee table control method according to claim 8, characterized in that: Redundant symbols include modal particles and repetition symbols. The word segmentation process is as follows: For the standardized dialect text, a word segmentation tool is used for processing. The word segmentation tool, combined with the collected dialect samples, uses a bidirectional maximum matching method to split the text, starting from the left and right sides of the text respectively. It prioritizes matching the longest valid words in the dictionary, and then compares the splitting results in the two directions, selecting the sequence with fewer words as the initial word segmentation result. For adjoining words not included in the dictionary, the basic morphemes contained in the adjoining words are identified, and the adjoining words are split into two or more words that can independently express part of the semantics. Each word obtained from word segmentation is labeled with its corresponding part of speech. The labeling is based on the conventional grammatical function of the word in the dialect, and the dialect word class library in the big data is referenced during the labeling. For core words related to instructions, special marking is performed to highlight their functions.

Images