A gas stove intelligent control method based on biometrics and AI learning

Abstract

The present invention discloses a gas stove intelligent control method based on biometrics and AI learning, which relates to the field of data processing technology; the method comprises: obtaining original audio data through an audio acquisition device, extracting current audio features from the original audio data, and verifying the user identity according to the current audio features; if the identity authentication is passed, parsing the current audio features into user text information and storing it; matching the context information of the current user text information; inputting the context information and the current user text information into a multi-round dialogue model to generate operation instructions and / or interactive text information; converting the interactive text information into interactive audio data and outputting it through an audio output device, and controlling the gas stove according to the operation instructions. The present invention combines the context information and the current user text information to generate operation instructions and / or interactive text information, so as to accurately identify the user's intention, realize intelligent decision-making and continuous interaction, thereby improving the user experience and reducing the risk of misoperation.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a gas stove intelligent control method based on biometrics and AI learning. Background Art

[0002] With the rapid development of smart home technology, modern users are increasingly inclined to use smart control devices to enhance the convenience and safety of daily life. As a basic and key kitchen facility, the demand for smart upgrades of gas stoves is becoming increasingly prominent.

[0003] At present, the control method of gas stoves mainly focuses on realizing basic operations of gas stoves, such as turning on and off and adjusting the fire power, through voice command control. This control method provides certain convenience for users and makes the operation simpler and more direct. However, when it comes to complex or continuous user voice input, the existing voice command control method often seems to be powerless. The reason is that the existing technology lacks effective context information integration capabilities, which makes it impossible to accurately understand the user's operation intention, and then cannot accurately understand and execute user requests, making it difficult to achieve a smooth user interaction experience. Summary of the invention

[0004] The present invention provides a gas stove intelligent control method based on biometrics and AI learning to solve at least one of the problems mentioned in the above background technology.

[0005] The specific technical solutions provided by this application are as follows:

[0006] A gas stove intelligent control method based on biometrics and AI learning, comprising the steps of:

[0007] Acquire raw audio data through an audio acquisition device, extract current audio features from the raw audio data, and verify the user's identity based on the current audio features;

[0008] If the identity authentication is successful, the current audio features are parsed into user text information and stored;

[0009] Matching the context information of the current user text information; the context information includes user text information, operation instructions and interactive text information;

[0010] The context information and the current user text information are input into a multi-round dialogue model to generate operation instructions and / or interactive text information; the interactive text information is converted into interactive audio data and output through an audio output device, and the gas stove is controlled according to the operation instructions.

[0011] As a preferred solution, the step of extracting the current audio features from the original audio data comprises the following steps:

[0012] The original audio signal is divided into several audio frames, and each audio frame is preprocessed to obtain corresponding spectrum data;

[0013] The spectrum data is passed through a Mel filter to obtain a Mel spectrum, a cepstrum analysis is performed based on the Mel spectrum to obtain a Mel frequency cepstrum coefficient, and the Mel frequency cepstrum coefficient is integrated into the current audio feature;

[0014] As a preferred solution, the preprocessing of each audio frame to obtain corresponding spectrum data comprises the steps of:

[0015] Acquire the operating status of the range hood in real time, and set the energy threshold and background noise power spectrum according to the operating status of the range hood;

[0016] Perform window processing on each audio frame and remove audio frames whose energy is lower than the energy threshold;

[0017] Obtain a power spectrum of each audio frame, and subtract the noise power spectrum from the power spectrum of the audio frame;

[0018] Perform fast Fourier transform on each audio frame to obtain spectrum data.

[0019] As a preferred solution, the method of verifying the user identity according to the current audio feature includes the following steps:

[0020] Input the current audio feature as a feature vector into the identity recognition model, and output the corresponding probability scores of several pre-stored user audio features;

[0021] If the probability score of the user's audio features and the current audio features is greater than the audio probability score threshold, the identity authentication is determined to be successful, and the user identity and the current audio features are recorded; otherwise, the verification fails.

[0022] As a preferred solution, the current audio feature is parsed into user text information and stored, comprising the steps of:

[0023] Convert the current audio features into a text sequence through a speech recognition model;

[0024] Perform keyword matching on the text sequence to obtain the operation keyword;

[0025] User text information that conforms to the preset storage structure is generated and stored based on the user identity, text sequence, operation keywords and data timestamp.

[0026] As a preferred solution, the method of converting the current audio features into a text sequence through a speech recognition model comprises the steps of:

[0027] Extracting accent features of the original audio data; the accent features include duration, pitch frequency and energy features of each phoneme;

[0028] Concatenating the accent feature with the current audio feature to obtain an updated current audio feature;

[0029] Input the current audio features into the speech recognition model to obtain the text sequence.

[0030] As a preferred solution, the multi-round dialogue model includes an encoder and a decoder; the encoder is constructed by a bidirectionally trained Transformer; and the decoder is constructed by a unidirectionally trained Transformer.

[0031] As a preferred solution, the step of inputting context information and current user text information into a multi-round dialogue model to generate operation instructions and / or interactive text information includes the following steps:

[0032] The encoder generates hidden layer representation based on context information and current user text information;

[0033] The decoder generates several operation instructions and their corresponding instruction confidences according to the hidden layer representation;

[0034] If the instruction confidence is lower than the first instruction confidence threshold, generating interactive text information for a clarification response;

[0035] If the command confidence is higher than the first command confidence threshold and lower than the second command confidence threshold, the operation command is transmitted to the gas stove controller, and an interactive text message of the spliced ​​operation description is generated for a confirmation reply;

[0036] If the command confidence is higher than the second command confidence threshold, the operation command is transmitted to the gas stove controller, and an interactive text message of an affirmative reply splicing operation description is generated;

[0037] The second instruction confidence threshold is greater than the first instruction confidence threshold.

[0038] As a preferred solution, before verifying the user identity according to the current audio feature, the method further includes the steps of: acquiring a facial image of the current user through an image acquisition device;

[0039] The method of verifying the user identity according to the current audio feature comprises the following steps:

[0040] Input the current audio feature as a feature vector into the identity recognition model, and output the corresponding probability scores of several pre-stored user audio features;

[0041] Input the current user's facial image into the facial recognition model and output probability scores corresponding to several pre-stored user facial features;

[0042] If the probability score of the user's audio features is greater than the audio probability score threshold, and the probability score of the user's facial features is greater than the facial probability score threshold, the identity authentication is determined to be successful, and the user identity and current audio features are recorded; otherwise, the verification fails.

[0043] As a preferred solution, the identity recognition model is constructed by the following steps:

[0044] Collect audio data samples of several users in different environments, and mark and classify them into legitimate users and illegitimate users;

[0045] Denoising and filtering the audio data samples, and extracting audio features from the audio data samples;

[0046] The identity recognition model is trained and verified using the loss function and Adam optimizer until the loss function converges or the verification accuracy reaches the expected value; the loss function is expressed as:

[0047]

[0048] Among them, L represents the loss function, N is the number of samples, is the true label of the i-th sample, The probability score of predicting the i-th sample as a legitimate user.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention verifies the user identity by extracting the current audio features, ensures that only authorized users can operate the gas stove, and improves the safety of the gas stove. The present invention parses the current audio features into user text information and stores it, inputs context information and the current user text information into a multi-round dialogue model to generate operation instructions and / or interactive text information, realizes user interaction according to the interactive text information, controls the gas stove according to the operation instructions, accurately identifies the user's intention, realizes intelligent decision-making and continuous interaction, thereby improving the user experience and reducing the risk of misoperation.

[0051] In the present embodiment, the multi-round dialogue model encodes context and user input, obtains confidence scores, and makes decisions based on confidence thresholds. It can not only process explicit instructions, but also ensure that the performed operations are in line with the user's intentions by generating information to be confirmed or clarified in ambiguous or uncertain situations, thereby achieving efficient and accurate interaction between users and devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0054] Figure 1 A schematic diagram of a flow chart of a gas stove intelligent control method based on biometrics and AI learning provided by an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of a process of preprocessing each audio frame to obtain corresponding spectrum data provided by an embodiment of the present invention;

[0056] Figure 3 A schematic diagram of a process for parsing current audio features into user text information and storing the information provided by an embodiment of the present invention;

[0057] Figure 4 A schematic diagram of a process of converting current audio features into a text sequence through a speech recognition model provided in an embodiment of the present invention;

[0058] Figure 5 A schematic diagram of a process for inputting context information and current user text information into a multi-round dialogue model to generate operation instructions and / or interactive text information provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0061] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0062] With the rapid development of smart home technology, modern users are increasingly inclined to use smart control devices to enhance the convenience and safety of daily life. As a basic and key kitchen facility, the demand for smart upgrades of gas stoves is becoming increasingly prominent.

[0063] At present, the intelligent control technology of gas stoves mainly focuses on realizing basic operations such as turning on and off and adjusting the fire power through voice commands. This control method undoubtedly provides users with a certain convenience and makes the operation simpler and more direct. However, when it comes to complex or continuous user voice input, the existing voice control system often seems to be unable to cope with it. The reason is that the existing technology lacks effective context information integration capabilities and cannot accurately understand the user's operation intentions, resulting in it being unable to accurately understand and execute user requests, making it difficult to achieve a smooth user interaction experience.

[0064] To solve the above problems, please refer to Figure 1 The present invention provides a gas stove intelligent control method based on biometrics and AI learning, comprising the steps of:

[0065] Acquire raw audio data through an audio acquisition device, extract current audio features from the raw audio data, and verify the user's identity based on the current audio features;

[0066] If the identity authentication is successful, the current audio features are parsed into user text information and stored;

[0067] Matching the context information of the current user text information; the context information includes user text information, operation instructions and interactive text information;

[0068] The context information and the current user text information are input into a multi-round dialogue model to generate operation instructions and / or interactive text information; the interactive text information is converted into interactive audio data and output through an audio output device, and the gas stove is controlled according to the operation instructions.

[0069] The present invention verifies the user identity by extracting the current audio features, ensures that only authorized users can operate the gas stove, and improves the safety of the gas stove. The present invention parses the current audio features into user text information and stores it, inputs context information and the current user text information into a multi-round dialogue model to generate operation instructions and / or interactive text information, realizes user interaction according to the interactive text information, controls the gas stove according to the operation instructions, accurately identifies the user's intention, realizes intelligent decision-making and continuous interaction, thereby improving the user experience and reducing the risk of misoperation.

[0070] Specifically, each step of the gas stove intelligent control method based on biometrics and AI learning of the present invention is described in detail through the following content:

[0071] A gas stove intelligent control method based on biometrics and AI learning, comprising the steps of:

[0072] S1. Obtaining original audio data through an audio acquisition device, extracting current audio features from the original audio data, and verifying the user's identity based on the current audio features;

[0073] Furthermore, the step of extracting the current audio features from the original audio data comprises the following steps:

[0074] S111, dividing the original audio signal into a plurality of audio frames, and preprocessing each audio frame to obtain corresponding spectrum data;

[0075] For further information, see Figure 2 , the preprocessing of each audio frame to obtain corresponding spectrum data comprises the steps of:

[0076] S1111, obtaining the operating state of the range hood in real time, and setting an energy threshold and a background noise power spectrum according to the operating state of the range hood;

[0077] The operating status of the range hood includes on, off, operating intensity level, etc. In most cases, the noise of the range hood is the most significant and persistent noise source in the kitchen environment, and its noise characteristics occupy a major part of the current ambient audio. In this embodiment, the operating status of the range hood and the corresponding energy threshold and background noise power spectrum are stored in a mapping table. By identifying the operating status of the range hood, the energy threshold and background noise power spectrum can be accurately set.

[0078] S1112, performing window processing on each audio frame, and removing audio frames whose energy is lower than an energy threshold;

[0079] The original audio signal is divided into several smaller audio frames, each of which has a fixed length. In order to obtain smoother and more accurate spectrum information, one embodiment performs windowing processing on each audio frame signal through a Hamming window, that is, a window function is applied to each frame to improve the signal processing effect by reducing the discontinuity of the frame boundary. The energy value of each audio frame is calculated, and the audio frames are screened according to the set energy threshold to filter out invalid or noise signals and retain the valid part.

[0080] S1113, obtaining a power spectrum of each audio frame, and subtracting a noise power spectrum from the power spectrum of the audio frame;

[0081] The noise power spectrum is subtracted from the power spectrum of each audio frame in order to reduce the impact of background noise on the audio signal and make the extracted spectrum data clearer and more accurate.

[0082] This embodiment simplifies the complexity of the noise reduction algorithm by focusing on the main noise sources. Compared with processing various uncertainties and frequency-varying environmental noises, focusing on processing the main noise sources makes the algorithm more efficient. It can be understood that for environmental noise with less impact, better results can be achieved by processing the background noise based on the range hood.

[0083] S1114. Perform fast Fourier transform on each audio frame to obtain spectrum data.

[0084] Perform a fast Fourier transform on each audio frame to convert the time domain signal into corresponding spectrum data to support the analysis of the energy distribution of the signal at different frequency components, thereby extracting frequency features useful for subsequent processing. The spectrum data obtained by the fast Fourier transform is expressed as:

[0085]

[0086] in, Represents the spectrum data of the frequency domain signal at the kth frequency index; n represents the sampling point index in the audio frame, and its value range is [0, N-1]; N is the length of the frame, that is, the number of sampling points in each audio frame; It represents the value of the windowed audio signal at the nth sampling point, where k is the frequency index.

[0087] S112, obtaining a Mel spectrum through a Mel filter on the spectrum data, performing a cepstrum analysis based on the Mel spectrum to obtain a Mel frequency cepstrum coefficient, and integrating the Mel frequency cepstrum coefficient into the current audio feature;

[0088] The Mel-frequency cepstral coefficients are expressed as:

[0089]

[0090] Where M represents the number of Mel filters. represents the mth Mel spectrum; Represents the nth Mel frequency cepstral coefficient value. The Mel frequency spectrum is represented as:

[0091]

[0092] in, represents the output of the mth Mel filter, that is, the mth Mel spectrum; Represents the power spectrum of the kth frequency index of the frequency domain signal; Represents the value of the mth Mel filter at frequency index k; k represents the frequency index.

[0093] Furthermore, the verification of the user identity according to the current audio feature comprises the steps of:

[0094] S121, inputting the current audio feature as a feature vector into the identity recognition model, and outputting probability scores corresponding to a number of pre-stored user audio features;

[0095] In one embodiment, the identity recognition model is based on a deep neural network (DNN), the input layer is audio features; the hidden layer uses 3 to 5 fully connected layers, each layer contains 128 to 256 neurons; output layer: the number of nodes in the output layer is the same as the number of registered users, which is used to reveal the probability that different users are the current users.

[0096] S122. If the probability score of the user's audio feature and the current audio feature is greater than the audio probability score threshold, it is determined that the identity authentication is passed, and the user identity and the current audio feature are recorded; otherwise, the authentication fails.

[0097] As a preferred embodiment, the identity recognition model is constructed by the following steps:

[0098] Collect audio data samples of several users in different environments and mark them; classify the audio data samples into legal users and illegal users;

[0099] Denoising and filtering the audio data samples, and extracting audio features from the audio data samples;

[0100] The identity recognition model is trained and verified using the loss function and Adam optimizer until the loss function converges or the verification accuracy reaches the expected value.

[0101] The loss function is expressed as:

[0102]

[0103] Among them, L represents the loss function, N is the number of samples, is the true label of the i-th sample (value is 0 or 1), The probability score of the i-th sample being predicted as a legitimate user. In this embodiment, the cross entropy loss is used as the loss function to evaluate the distance between the predicted probability distribution output by the model and the true distribution of the actual label.

[0104] S2. If the identity authentication is passed, the current audio features are parsed into user text information and stored;

[0105] See also Figure 3 , the current audio feature is parsed into user text information and stored, including the steps of:

[0106] S21, converting the current audio features into a text sequence through a speech recognition model;

[0107] The speech recognition model is based on a bidirectional recurrent neural network model (Bi-RNN). The input layer is the current audio features, which at least include the Mel-frequency cepstral coefficients obtained in the previous step; the hidden layer includes two independent RNN layers, denoted as the forward layer and the backward layer. The forward layer processes the time series of audio data, traversing the audio frames from front to back; the backward layer processes the data from back to front. During the speech recognition process, the speech recognition model obtains the current audio features through the input layer, and the forward layer and the backward layer traverse the audio frames from front to back and from back to front, respectively, to convert the current audio features of the complex time series pattern into the corresponding text sequence.

[0108] S22, performing keyword matching on the text sequence to obtain an operation keyword;

[0109] Using natural language processing (NLP) technology, the text sequence is segmented and annotated to extract the operation keywords. Keyword matching uses a dictionary matching algorithm or an NLP-based keyword extraction algorithm to ensure accurate recognition of user intent, such as "turn on", "turn off", "increase", "time", "timing" and other operation commands.

[0110] S23. Generate and store user text information that complies with a preset storage structure according to the user identity, text sequence, operation keyword and data timestamp.

[0111] In one embodiment, the storage structure includes fields: user identification, text sequence, operation keyword, data timestamp. A unified data format such as JSON format is used in the storage structure to store each parsed voice conversion information in a data table.

[0112] In one embodiment, please refer to Figure 4 , the current audio features are converted into a text sequence through a speech recognition model, including the steps of:

[0113] S211, extracting accent features of the original audio data; the accent features include duration, pitch frequency and energy features of each phoneme;

[0114] Based on the audio frames obtained in step S111 "denoising and filtering the original audio data, dividing the original audio signal into several audio frames", this step S211 uses a pre-trained automatic speech recognition (ASR) model to map the audio frames to phonemes, calculate the duration of each phoneme, and the oral morphology affects the phoneme length of different accents. For example, some phonemes may last longer in certain accents; measure the pitch frequency characteristics of the audio frames, that is, calculate the statistical data such as the mean, standard deviation, range and change rate of the pitch. Some accents have specific pitch frequency characteristics; measure the energy characteristics of the audio frames, such as the average energy, peak energy and energy distribution characteristics. Some accents have specific energy characteristics on specific phonemes. The duration, pitch frequency and energy characteristics of the extracted phonemes are combined to form a comprehensive feature vector, which is recorded as the accent feature.

[0115] S212, concatenating the accent feature with the current audio feature to obtain an updated current audio feature;

[0116] The extracted accent features (including duration, pitch frequency features and energy features) are concatenated with the current audio frame features (including Mel-frequency cepstral coefficients) to obtain updated current audio features; the updated current audio features will contain more comprehensive audio feature information.

[0117] S213: Input the current audio features into the speech recognition model to obtain a text sequence. The speech recognition model obtains the current audio features through the input layer, and the forward layer and the backward layer traverse the audio frames from front to back and from back to front respectively, and converts the current audio features of the complex time series pattern into the corresponding text sequence.

[0118] This embodiment combines accent features with current audio features and uses a speech recognition model to convert the audio features into accurate text sequences, thereby improving the adaptability of the speech recognition model to specific accents and enhancing the accuracy and robustness of speech recognition.

[0119] S3, matching the context information of the current user text information; the context information includes user text information, operation instructions and interactive text information;

[0120] Match the context information of the current user text information, specifically matching the user text information, operation instructions and interactive text information according to the user identity. The user text information is all the user text information of the current round of dialogue stored each time the aforementioned step S23 is executed, and each user text information includes the user identity, text sequence, operation keywords and data timestamp. The operation instruction is the operation instruction in the historical record of the dialogue between the multi-round dialogue model and the current user. The interactive text information is the interactive text information in the historical record of the dialogue between the multi-round dialogue model and the current user.

[0121] S4. Input the context information and the current user text information into a multi-round dialogue model to generate operation instructions and / or interactive text information; convert the interactive text information into interactive audio data through an audio output device and output it, and control the gas stove according to the operation instructions.

[0122] The multi-round dialogue model includes an encoder and a decoder. The encoder is constructed by a bidirectionally trained Transformer, which can consider the context information of a word on its left and right sides, so that it has a stronger semantic representation ability when understanding words in a sentence; the decoder is constructed by a unidirectionally trained Transformer, which only considers the context information from left to right, which makes it more suitable for generation tasks. In the process of inputting context information and current user text information into the multi-round dialogue model to generate operation instructions and / or interactive text information, the encoder processes the context information and the current user text information through word segmentation and word embedding to obtain a word embedding sequence, and generates a hidden layer representation according to the word embedding sequence; the decoder generates an operation instruction and its corresponding instruction confidence according to the hidden layer representation, such as "set_flame_level(5),"confidence": 0.92" indicates that the operation instruction "adjust the firepower to the fifth level firepower" and the instruction confidence is 0.92.

[0123] Among them, context information is used to understand the current user's intention and maintain the continuity and consistency of the dialogue. For example, if the user previously instructed the gas stove to "turn on the medium heat", the system needs to remember this operation so that the user does not have to repeat it in subsequent interactions; the current user text information represents the user's current operation requirements or inquiry content, which serves as the core input of this interaction; the historical context information and the current user text information are combined as comprehensive input to ensure that the multi-round dialogue model can understand and generate based on the current and past interaction contexts. After the multi-round dialogue model parses the context information and the current user text information, it generates operation instructions and / or interaction text information. For example, the operation instruction is to adjust the fire level of the gas stove; in addition to the operation instruction, the multi-round dialogue model may also generate interaction text information to continue interacting with the user. For example, the model may generate "The gas stove has been adjusted to medium heat" to feedback the operation results, or generate "Do you need any other help?" to guide further interaction.

[0124] By matching the current user's text information with context information, the system can accurately identify and understand the user's intentions, ensure the continuity and consistency of the conversation, thereby avoiding erroneous operations due to incomplete understanding of individual instructions and ensuring high accuracy of device execution.

[0125] As a preferred embodiment, please refer to Figure 5 The step of inputting the context information and the current user text information into the multi-round dialogue model to generate operation instructions and / or interactive text information includes the following steps:

[0126] The encoder generates hidden layer representation based on context information and current user text information;

[0127] The decoder generates several operation instructions and their corresponding instruction confidences according to the hidden layer representation;

[0128] If the instruction confidence is lower than the first instruction confidence threshold, an interactive text message of a reply to be clarified is generated; for example, the interactive text message of the reply to be clarified is “Sorry, I don’t quite understand, can you explain it in more detail?”.

[0129] If the command confidence is higher than the first command confidence threshold and lower than the second command confidence threshold, the operation command is transmitted to the gas stove controller, and an interactive text message of the spliced ​​operation description is generated for a confirmation reply;

[0130] If the command confidence is higher than the second command confidence threshold, the operation command is transmitted to the gas stove controller to generate an interactive text message of an affirmative reply splicing operation description.

[0131] The second instruction confidence threshold is greater than the first instruction confidence threshold.

[0132] As an example, firepower is quantified and stored as a control quantity through the first to tenth levels; in a dialogue round, the current user text information "turn the fire to the highest, and turn it down after ten minutes" is parsed by the multi-round dialogue model into two operation instructions: "adjust to the first level of firepower immediately" and "adjust to the tenth level of firepower after ten minutes."

[0133] Among them, the multi-round dialogue model selects the operation instruction "immediately adjust the fire to the first level" with the highest command confidence based on "turn the fire to the highest", and if it is judged that the command confidence is higher than the second command confidence threshold, the operation instruction is transmitted to the gas stove controller; and an interactive text message with an affirmative reply splicing the operation description is generated, such as "OK, the fire has been turned to the highest level."

[0134] The user text message "Turn down the fire in ten minutes" is difficult to quantify into specific instructions. The multi-round dialogue model selects the instruction with the highest confidence level "Turn down the fire in ten minutes" based on "Turn down the fire in ten minutes", and judges that the instruction confidence level is higher than the first instruction confidence threshold and lower than the second instruction confidence threshold. Then, the operation instruction is transmitted to the gas stove controller, and an interactive text message with a reply to be confirmed and the operation description is generated, such as "OK, the fire will be adjusted to the lowest level in ten minutes, please confirm".

[0135] After receiving the interactive audio, the user supplements and updates the current user text information "No need to adjust it to the lowest setting, adjust it to medium heat". Based on the context information matching analysis, the multi-round dialogue model understands the user's intention to adjust the fire to the fifth level after ten minutes. The multi-round dialogue model outputs the operation instruction with the highest command confidence "Adjust the fire to the fifth level after ten minutes", and if the command confidence is higher than the second command confidence threshold, the operation instruction is transmitted to the gas stove controller; an interactive text message with an affirmative reply splicing the operation description is generated, such as "OK, adjust the fire to medium heat after ten minutes".

[0136] The multi-round dialogue model in this embodiment can not only process explicit instructions by encoding context and user input, obtaining confidence scores, and making decisions based on confidence thresholds, but also ensure that the performed operations are in line with the user's intentions by generating information to be confirmed or clarified in ambiguous or uncertain situations, thereby achieving efficient and accurate interaction between users and devices.

[0137] As a preferred embodiment, in step S1, before verifying the user identity according to the current audio feature, it also includes the steps of: acquiring a facial image of the current user through an image acquisition device;

[0138] The method of verifying the user identity according to the current audio feature comprises the following steps:

[0139] Input the current audio feature as a feature vector into the identity recognition model, and output the corresponding probability scores of several pre-stored user audio features;

[0140] Input the current user's facial image into the facial recognition model and output probability scores corresponding to several pre-stored user facial features;

[0141] If the probability score of the user's audio features is greater than the audio probability score threshold, and the probability score of the user's facial features is greater than the facial probability score threshold, the identity authentication is determined to be successful, and the user identity and current audio features are recorded; otherwise, the verification fails.

[0142] This embodiment is different from the previous embodiment in that, in addition to obtaining the original audio data through the audio acquisition device, it also adds obtaining the user's facial image through the image acquisition device. By combining the audio features extracted from the audio data and the facial features extracted from the image data, the user's identity is doubly verified, thereby improving the accuracy and security of identity recognition.

[0143] A gas stove intelligent control system based on biometrics and AI learning, comprising:

[0144] Audio acquisition module, used to obtain original audio data;

[0145] An audio parsing module is used to extract current audio features from raw audio data, and parse the current audio features into user text information and store it;

[0146] An authentication module, used to verify the user's identity based on the current audio features;

[0147] A dialogue interaction module, used to input context information and current user text information into a multi-round dialogue model to generate operation instructions and / or interactive text information;

[0148] An audio output processing module, used to convert interactive text information into interactive audio data and output it;

[0149] The gas stove control module is used to control the gas stove according to operating instructions.

[0150] In the embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the modules can be electrical, mechanical or other forms.

[0151] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0152] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.

[0153] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

[0154] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A gas stove intelligent control method based on biometrics and AI learning, characterized by: Includes steps: Acquire raw audio data through an audio acquisition device, extract current audio features from the raw audio data, and verify the user's identity based on the current audio features; If the identity authentication is successful, the current audio features are parsed into user text information and stored; Matching the context information of the current user text information; the context information includes user text information, operation instructions and interactive text information; Input context information and current user text information into a multi-round dialogue model to generate operation instructions and / or interactive text information; convert the interactive text information into interactive audio data through an audio output device and output it, and control the gas stove according to the operation instructions; The step of inputting the context information and the current user text information into the multi-round dialogue model to generate operation instructions and / or interactive text information includes the following steps: The encoder generates hidden layer representation based on context information and current user text information; The decoder generates several operation instructions and their corresponding instruction confidences according to the hidden layer representation; If the instruction confidence is lower than the first instruction confidence threshold, generating interactive text information for a clarification response; If the command confidence is higher than the first command confidence threshold and lower than the second command confidence threshold, the operation command is transmitted to the gas stove controller, and an interactive text message of the spliced ​​operation description is generated for a confirmation reply; If the command confidence is higher than the second command confidence threshold, the operation command is transmitted to the gas stove controller, and an interactive text message of an affirmative reply splicing operation description is generated; The second instruction confidence threshold is greater than the first instruction confidence threshold; The method of extracting the current audio features from the original audio data comprises the following steps: The original audio signal is divided into several audio frames, and each audio frame is preprocessed to obtain corresponding spectrum data; The spectrum data is passed through a Mel filter to obtain a Mel spectrum, a cepstrum analysis is performed based on the Mel spectrum to obtain a Mel frequency cepstrum coefficient, and the Mel frequency cepstrum coefficient is integrated into the current audio feature; The method of verifying the user identity according to the current audio feature comprises the following steps: Input the current audio feature as a feature vector into the identity recognition model, and output the corresponding probability scores of several pre-stored user audio features; If the probability score of the user's audio feature and the current audio feature is greater than the audio probability score threshold, the identity authentication is determined to be successful, and the user's identity and the current audio feature are recorded; otherwise, the authentication fails; The step of parsing the current audio features into user text information and storing the information comprises the following steps: Convert the current audio features into a text sequence through a speech recognition model; Perform keyword matching on the text sequence to obtain the operation keyword; User text information that conforms to the preset storage structure is generated and stored based on the user identity, text sequence, operation keywords and data timestamp.

2. According to claim 1, a gas stove intelligent control method based on biometrics and AI learning is characterized by: The preprocessing of each audio frame to obtain corresponding spectrum data comprises the steps of: Acquire the operating status of the range hood in real time, and set the energy threshold and background noise power spectrum according to the operating status of the range hood; Perform window processing on each audio frame and remove audio frames whose energy is lower than the energy threshold; Obtain a power spectrum of each audio frame, and subtract the noise power spectrum from the power spectrum of the audio frame; Perform fast Fourier transform on each audio frame to obtain spectrum data.

3. According to claim 1, a gas stove intelligent control method based on biometrics and AI learning is characterized by: The method of converting the current audio features into a text sequence through a speech recognition model comprises the following steps: Extracting accent features of the original audio data; the accent features include duration, pitch frequency and energy features of each phoneme; Concatenating the accent feature with the current audio feature to obtain an updated current audio feature; Input the current audio features into the speech recognition model to obtain the text sequence.

4. According to claim 1, a gas stove intelligent control method based on biometrics and AI learning is characterized by: The multi-round dialogue model includes an encoder and a decoder; the encoder is constructed by a bidirectionally trained Transformer; and the decoder is constructed by a unidirectionally trained Transformer.

5. The method for intelligently controlling a gas stove based on biometrics and AI learning according to claim 1 is characterized in that: Before verifying the user identity according to the current audio feature, the method further includes the steps of: acquiring a facial image of the current user through an image acquisition device; The method of verifying the user identity according to the current audio feature comprises the following steps: Input the current audio feature as a feature vector into the identity recognition model, and output the corresponding probability scores of several pre-stored user audio features; Input the current user's facial image into the facial recognition model and output probability scores corresponding to several pre-stored user facial features; If the probability score of the user's audio features is greater than the audio probability score threshold, and the probability score of the user's facial features is greater than the facial probability score threshold, the identity authentication is determined to be successful, and the user identity and current audio features are recorded; otherwise, the verification fails.

6. The gas stove intelligent control method based on biometrics and AI learning according to claim 1 is characterized in that: The identity recognition model is constructed by the following steps: Collect audio data samples of several users in different environments, and mark and classify them into legitimate users and illegitimate users; Denoising and filtering the audio data samples, and extracting audio features from the audio data samples; The identity recognition model is trained and verified using the loss function and Adam optimizer until the loss function converges or the verification accuracy reaches the expected value; the loss function is expressed as: ； in, represents the loss function, is the sample size, is the true label of the i-th sample, The probability score of predicting the i-th sample as a legitimate user.

Images