Intelligent voice interaction method and system of osteoporosis therapeutic apparatus

Abstract

The invention relates to the technical field of intelligent voice interaction, discloses an intelligent voice interaction method and system for an osteoporosis therapeutic instrument, and aims to improve the convenience of equipment and the user experience. According to the method, user voice interaction data is acquired, a user intention is identified by using an intention identification model, and a therapeutic apparatus response strategy is generated based on a dialogue management model constructed by a deep reinforcement learning A3C algorithm. The system converts the response strategy into voice output, and provides efficient and intelligent interaction experience for the user. The dialogue management model is trained by adopting a dialogue data set containing multiple rounds of dialogue scenes, so that the intention of the user is accurately understood and proper response is given. Through intelligent voice interaction, the user can easily know information such as the treatment principle, the operation process and the treatment effect, and the usability of the equipment and the user satisfaction degree are remarkably improved. According to the invention, the operation process of the osteoporosis therapeutic instrument is simplified, and the intelligent level of voice interaction is improved through the deep reinforcement learning technology.

Description

Technical Field

[0001] The present invention relates to the field of intelligent voice interaction technology, and specifically to an intelligent voice interaction method and system for an osteoporosis therapeutic device. Background Art

[0002] With the rapid development of medical technology, the intelligence and convenience of various medical devices and systems have become an important trend in the development of the medical industry. As a treatment device for osteoporosis, osteoporosis therapeutic devices have a wide range of applications in the medical field. However, traditional osteoporosis therapeutic devices have certain complexities and limitations in operation and use, and often require professional operation by medical staff, which to a certain extent limits the convenience and popularity of the device.

[0003] In order to improve user experience and the convenience of device use, intelligent voice interaction technology has gradually been introduced into medical devices. Intelligent voice interaction technology allows users to interact with devices through natural language, thereby simplifying the operation process and improving the usability of the device. However, existing voice interaction systems still face challenges in handling complex multi-round conversations, understanding user intent, and generating appropriate responses.

[0004] Especially in the application scenario of osteoporosis treatment device, users may need to understand the treatment principle, operation process, treatment effect and other information, which requires the voice interaction system to accurately understand the user's various intentions and give appropriate responses. Due to the particularity of medical scenarios, the quality of dialogue, user satisfaction and task completion are all important indicators for evaluating the performance of voice interaction systems. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent voice interaction method and system for an osteoporosis therapeutic device to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: an intelligent voice interaction method for an osteoporosis therapeutic device, the method comprising: Acquire the user's voice interaction data, perform feature extraction on the voice data, and obtain a voice feature vector; Input the speech feature vector into the intention recognition model to obtain the user intention, and input the user intention into the dialogue management model to obtain the therapeutic device response strategy; Post-processing the therapeutic device response strategy to obtain a therapeutic device response instruction, further converting the therapeutic device response instruction into a voice output to generate a final therapeutic device response information; The dialogue management model is constructed based on the deep reinforcement learning A3C algorithm, and a dialogue data set containing multiple dialogue scenarios is constructed as training data; wherein each dialogue scenario contains a user voice input sequence, a therapeutic device response sequence and a corresponding reward value; the user voice input sequence records each round of voice input information of the user during the dialogue process, and the therapeutic device response sequence records each round of response information given by the therapeutic device to the user input, and the reward value is set according to the quality of the dialogue, user satisfaction and task completion.

[0007] Preferably, the step of training the dialogue management model includes: Step 1: Initialize the dialogue management model based on asynchronous advantage actor-critic A3C, define the state space of the model as the context information of the current dialogue, which includes the historical voice information of the user input, the historical information of the therapeutic device response, and the current state of the dialogue; define the action space of the model as the optional set of therapeutic device responses, which includes all response types or response templates given by the therapeutic device; define the reward function of the model as the dialogue quality score, which is used to calculate and distribute rewards during the training process; Step 2: Use the constructed dialogue dataset to train the dialogue management model. During the training process, an asynchronous update method is used for exploration and utilization. By continuously interacting with the environment, samples of states, actions, rewards, and transitions to the next state are collected to optimize the model’s decision-making strategy. Step 3: During the training process, calculate the average reward value of the model in the simulated dialogue; when the average reward value reaches the preset level, the training is completed and a trained dialogue management model is obtained.

[0008] Preferably, the method of extracting features from speech data to obtain a speech feature vector comprises: Collect voice data of the user interacting with the therapeutic device; Preprocessing the collected speech data to obtain a preprocessed speech data set; the preprocessing step includes noise reduction and frame processing; The preprocessed speech data set is input into a speech processing model, wherein the input of the speech processing model is speech data and the output is a speech feature vector.

[0009] Preferably, the training step of the speech processing model includes: Select the convolutional recurrent neural network CRNN model as the basic architecture of the speech processing model; Acquire speech training data, wherein the speech training data includes speech samples and their corresponding speech recognition labels; The CRNN model is trained using speech training data, and the CTC loss function is used for optimization during the training process; When the recognition accuracy of the speech processing model reaches a preset threshold, the training is stopped to obtain a trained speech processing model.

[0010] Preferably, the training step of the intent recognition model includes: Select the long short-term memory network LSTM model as the basic architecture of the intent recognition model; Acquire intent training data, wherein the intent training data includes a speech feature vector and a corresponding user intent label; The LSTM model is trained using the intent training data, and the cross entropy loss function is used for optimization during the training process; When the accuracy of the intent recognition model reaches a preset threshold, the training is stopped to obtain a trained intent recognition model.

[0011] Preferably, in the A3C algorithm, the actor network is used to generate action probability distribution, and the critic network is used to estimate the state value function; during the training process, the actor network selects actions according to the current strategy, and the critic network evaluates the value of the action, and the actor network and the critic network are jointly optimized through the policy gradient and value function approximation methods.

[0012] Preferably, the user voice input sequence and the therapeutic device response sequence are converted into text form through speech recognition and text conversion technology for training and reasoning of the dialogue management model.

[0013] Preferably, the step of converting the therapeutic device response instruction into voice output comprises: The end-to-end text-to-speech synthesis model Tacotron is selected as the basic model. Its input is the text form of the therapeutic device response instructions, and its output is the synthesized speech signal. The therapeutic device response instructions are input into the text-to-speech synthesis model to generate the final speech response information.

[0014] Preferably, the step of post-processing the therapeutic device response strategy includes: Perform grammatical verification on the therapy device response strategy; Conducting contextual consistency checks on therapy device response strategies.

[0015] Preferably, an intelligent voice interaction system for an osteoporosis therapeutic apparatus comprises: A voice data acquisition module is used to acquire the user's voice interaction data; A feature extraction module, connected to the voice data acquisition module, for extracting features from the acquired voice data to obtain a voice feature vector; An intention recognition module, connected to the feature extraction module, for inputting the speech feature vector into the intention recognition model to obtain the user intention; A dialogue management module, connected to the intention recognition module, constructed based on the deep reinforcement learning A3C algorithm, and used to input the user's intention into the dialogue management model to obtain the therapeutic device response strategy, wherein the training data of the dialogue management model includes a dialogue data set of constructed multi-round dialogue scenarios, each dialogue scenario includes a user voice input sequence, a therapeutic device response sequence, and a corresponding reward value; A post-processing module, connected to the dialogue management module, for post-processing the therapeutic device response strategy to obtain a therapeutic device response instruction; The voice output module is connected to the post-processing module and is used to convert the therapeutic device response instruction into voice output to generate the final therapeutic device response information. Compared with the prior art, the present invention has the following beneficial effects: By introducing intelligent voice interaction technology, users can interact with the osteoporosis treatment device through natural language without the need for professional medical care, which greatly reduces the threshold for use. This interactive method is more intuitive and convenient, significantly improving the user experience. Traditional osteoporosis treatment devices are complicated to operate, and the present invention simplifies the operation process through intelligent voice interaction methods. Users only need to state their needs or questions, and the system can quickly understand and give corresponding responses, greatly enhancing the ease of use of the device.

[0016] The present invention adopts a dialogue management model constructed by a deep reinforcement learning A3C algorithm, which can accurately and efficiently understand the user's various intentions, including consultation on treatment principles, inquiry on operation procedures, understanding of treatment effects, etc., so as to provide users with more accurate information and services. By constructing a dialogue data set containing multiple rounds of dialogue scenarios and combining the reward value mechanism for model training, the dialogue management model can generate more appropriate and natural responses. This not only improves the quality of the dialogue, but also further improves user satisfaction. The implementation of the present invention provides new ideas and methods for the intelligent development of medical equipment. By combining advanced deep learning and reinforcement learning technologies, more medical equipment can be promoted to be intelligent and convenient, so as to better serve the majority of patients and medical staff. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a working principle diagram of the intelligent voice interaction method of the present invention; Figure 2 Diagram of the steps for training a dialogue management model; Figure 3 Diagram of the steps to train an intent recognition model. DETAILED DESCRIPTION

[0018] 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.

[0019] See also Figure 1-3 The present invention provides a technical solution: an intelligent voice interaction method for an osteoporosis therapeutic device, the method comprising: Step 1: Obtain the user's voice interaction data.

[0020] When using the osteoporosis treatment device, users can input interactive information in the form of voice. The system first obtains the user's voice interaction data through a microphone or other voice collection device.

[0021] Step 2: Extract features from speech data.

[0022] After obtaining the user's voice interaction data, the system needs to preprocess and extract features from the data. Preprocessing includes operations such as denoising and framing, and feature extraction can use methods such as MFCC (Mel Frequency Cepstral Coefficient) to obtain voice feature vectors.

[0023] Step 3: Identify user intent.

[0024] The speech feature vector obtained in step 2 is input into a pre-trained intention recognition model. The model can use a classification model in deep learning to recognize the user's intention and obtain the user's specific intention.

[0025] Step 4: Generate a therapeutic device response strategy.

[0026] The user intent identified in step 3 is input into the dialogue management model, which is built based on the deep reinforcement learning A3C algorithm. The dialogue management model generates the response strategy of the therapeutic device based on the current user intent and historical dialogue information. During this process, the model refers to a pre-built dialogue dataset containing multiple rounds of dialogue scenarios. The dataset contains a rich user voice input sequence, therapeutic device response sequence, and corresponding reward values, providing a learning basis for the model.

[0027] Step 5: Post-process and generate final response information.

[0028] The therapeutic device response strategy generated in step 4 is post-processed, such as text generation, grammar correction, etc., to obtain a therapeutic device response instruction. Further, the therapeutic device response instruction is converted into a voice output and played to the user through a speaker or other device to generate the final therapeutic device response information.

[0029] The present invention will be further described below in conjunction with Examples 1 to 3: Embodiment 1: The intelligent voice interaction method proposed in the present invention mainly involves the training of a dialogue management model, the feature extraction of voice data, and the application of the A3C algorithm. The implementation of these steps will be described in detail below with reference to specific examples.

[0030] The training steps of the dialogue management model include: Step 1: Initialize the dialogue management model The dialogue management model is initialized based on the asynchronous advantage actor-critic (A3C) algorithm. The state space of the model is defined as the context information of the current dialogue, including the historical voice information of the user input, the historical information of the therapeutic device response, and the current state of the dialogue. This information provides rich context for the model and helps to understand the user's intention more accurately.

[0031] The action space is defined as a set of optional therapeutic device responses, including all response types or response templates given by the therapeutic device. For example, when a user asks "How do I use this therapeutic device?", possible response templates include "You need to press the power button first, and then select the treatment mode that suits you." etc.

[0032] The reward function is defined as a dialogue quality score that is given based on factors such as dialogue fluency, accuracy, and user satisfaction. During training, the reward function will be used to calculate and distribute rewards to guide the model to learn better dialogue strategies.

[0033] Step 2: Training using the conversation dataset The dialogue management model is trained using a dialogue dataset that contains multi-round dialogue scenarios. During the training process, an asynchronous update method is used for exploration and utilization, that is, multiple training threads are carried out in parallel, and each thread interacts with the environment to collect state, action, reward, and next state transition samples.

[0034] By continuously optimizing the model’s decision-making strategy, the model can select appropriate responses based on different conversation contexts. For example, when a user asks about the effectiveness of treatment, the model can accurately understand the user’s intention and give an appropriate answer.

[0035] Step 3: Calculate the average reward value and determine whether the training is complete During the training process, the average reward value of the model in the simulated dialogue is continuously calculated. When the value reaches the preset level, it means that the model has learned a good dialogue strategy. At this time, the training is completed and a trained dialogue management model is obtained.

[0036] Methods for extracting features from speech data include: Collect the voice data of the user interacting with the therapeutic device. These data are original, unprocessed voice signals. Preprocess the collected voice data. The preprocessing steps include noise reduction and frame processing. Noise reduction aims to remove background noise and interference signals in the voice signal to improve the clarity and recognizability of the voice. Frame processing is to cut the continuous voice signal into multiple short frames to facilitate subsequent feature extraction and analysis.

[0037] The preprocessed speech data set is input into the speech processing model. The model can use structures such as convolutional neural network (CNN) or recurrent neural network (RNN) in deep learning to extract effective feature vectors from speech data. These feature vectors will serve as the basic input for subsequent intent recognition and dialogue management.

[0038] In the A3C algorithm, the actor network is used to generate action probability distribution, that is, the probability of selecting each action based on the current state; while the critic network is used to estimate the state value function, that is, to evaluate the quality of the current state and the possible rewards in the future.

[0039] During the training process, the actor network selects actions based on the current strategy and interacts with the environment; the critic network evaluates the value of the action and gives a reward prediction. The actor network and the critic network are jointly optimized through the policy gradient and value function approximation methods, so that the model can learn better action selection strategies and value evaluation capabilities.

[0040] For example, when a user asks "Can this therapeutic device treat osteoporosis?" during the interaction between the user and the therapeutic device, the actor network will generate a probability distribution of possible response actions based on the current dialogue context, while the critic network will evaluate the value of these responses and give reward predictions. By continuously optimizing and adjusting the model parameters, the therapeutic device can give accurate and user-satisfactory responses.

[0041] Embodiment 2: The training steps of the speech processing model include: Step 1: Select the convolutional recurrent neural network (CRNN) as the basic architecture of the speech processing model. CRNN combines the advantages of convolutional neural network (CNN) and recurrent neural network (RNN) and can effectively process sequence data, especially speech signals, which have time series characteristics.

[0042] Step 2: Obtain a large amount of speech training data, which includes speech samples and their corresponding speech recognition labels. Speech samples can be speech recordings from different speakers with different speaking speeds and intonations, while speech recognition labels are the text content corresponding to these speech samples.

[0043] Step 3: Use the acquired speech training data to train the CRNN model. During the training process, the connection temporal classification (CTC) loss function is used for optimization. The CTC loss function can handle the alignment problem in sequence data, allowing the model to learn speech recognition tasks without the need for precise alignment.

[0044] By continuously optimizing the model parameters, the CRNN model can accurately convert speech signals into corresponding text content.

[0045] Step 4: During the training process, the recognition accuracy of the speech processing model is continuously evaluated. When the recognition accuracy reaches the preset threshold, it means that the model has sufficient recognition ability. At this time, the training is stopped and a trained speech processing model is obtained.

[0046] The training steps of the intent recognition model include: Step 1: Select the long short-term memory network (LSTM) as the basic architecture of the intent recognition model. LSTM is a special RNN structure that can effectively handle long-term dependencies in long sequence data and is suitable for tasks such as intent recognition that require understanding contextual information.

[0047] Step 2: Obtain a large amount of intent training data, which includes speech feature vectors and their corresponding user intent labels. Speech feature vectors can be extracted from speech samples using the previously trained speech processing model, while user intent labels are the user intent categories corresponding to these speech samples.

[0048] Step 3: Use the acquired intent training data to train the LSTM model. During the training process, the cross entropy loss function is used for optimization. The cross entropy loss function can measure the difference between the probability distribution predicted by the model and the true label, thereby guiding the model to learn better intent recognition capabilities.

[0049] By continuously optimizing the model parameters, the LSTM model can accurately identify the user's intention based on the speech feature vector.

[0050] Step 4: During the training process, the accuracy of the intent recognition model is continuously evaluated. When the accuracy reaches the preset threshold, it means that the model has sufficient intent recognition capabilities. At this time, the training is stopped and a trained intent recognition model is obtained.

[0051] Embodiment 3: When the user interacts with the therapeutic device through voice, the user's voice input and the therapeutic device's voice response need to be converted into text form through speech recognition and text conversion technology. This step is crucial for the subsequent dialogue management model training and reasoning because it ensures the accuracy and processability of the interactive information.

[0052] In specific implementation, a mature speech recognition engine, such as an automatic speech recognition (ASR) system based on deep learning, can be used to convert user voice input into text. Similarly, the voice response of the therapeutic device can also be converted through the same ASR system or a dedicated text conversion tool. These converted text data will be used in the training and reasoning process of the dialogue management model to improve the performance and accuracy of the model.

[0053] In order to output the response instructions of the therapeutic device to the user in the form of voice, the present invention adopts the end-to-end text-to-speech synthesis model Tacotron as the basic model. The Tacotron model can receive text input and directly generate the corresponding voice signal, thereby realizing the conversion from text to speech. The specific implementation steps are as follows: Select and configure the Tacotron model: Select the appropriate Tacotron model version according to actual needs, and make necessary configurations and adjustments to ensure that the generated voice quality and effect meet the requirements.

[0054] Input therapeutic device response instructions: Input the therapeutic device response instructions (in text form) processed by the dialogue management model into the Tacotron model. These instructions may include treatment suggestions, operation instructions, precautions, etc.

[0055] Generate voice response information: After receiving text input, the Tacotron model will use its internal deep learning structure and algorithm to generate the corresponding voice signal. This process includes steps such as text encoding, acoustic feature prediction and waveform synthesis, and finally generates voice response information that the user can hear.

[0056] In order to ensure the consistency and accuracy of the response strategy of the therapeutic device in terms of syntax and context, the present invention also includes a step of post-processing the response strategy. The specific implementation is as follows: Syntax check: Use grammar checking tools or algorithms in natural language processing (NLP) technology to perform grammar check on the response strategy of the therapeutic device. This can help discover and correct potential grammatical errors or non-compliant expressions, thereby improving the accuracy and readability of the response.

[0057] Context consistency check: Combine the context information of the conversation to check the consistency of the response strategy of the therapeutic device. This is mainly to ensure that the response content maintains logical coherence and consistency with the previous conversation content, and avoid abrupt or contradictory statements. If inconsistencies are found, the system will make corresponding adjustments or corrections.

[0058] The present invention also includes an intelligent voice interaction system for an osteoporosis therapeutic apparatus, the system comprising: A voice data acquisition module is used to acquire the user's voice interaction data; A feature extraction module, connected to the voice data acquisition module, for extracting features from the acquired voice data to obtain a voice feature vector; An intention recognition module, connected to the feature extraction module, for inputting the speech feature vector into the intention recognition model to obtain the user intention; A dialogue management module, connected to the intention recognition module, constructed based on the deep reinforcement learning A3C algorithm, and used to input the user's intention into the dialogue management model to obtain the therapeutic device response strategy, wherein the training data of the dialogue management model includes a dialogue data set of constructed multi-round dialogue scenarios, each dialogue scenario includes a user voice input sequence, a therapeutic device response sequence, and a corresponding reward value; A post-processing module, connected to the dialogue management module, for post-processing the therapeutic device response strategy to obtain a therapeutic device response instruction; The voice output module is connected to the post-processing module and is used to convert the therapeutic device response instruction into a voice output to generate the final therapeutic device response information.

[0059] The implementation of the system refers to the above embodiment and will not be described in detail in the specification.

[0060] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0061] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent voice interaction method for osteoporosis therapeutic apparatus, characterized in that: The method comprises: Acquire the user's voice interaction data, perform feature extraction on the voice data, and obtain a voice feature vector; Input the speech feature vector into the intention recognition model to obtain the user intention, and input the user intention into the dialogue management model to obtain the therapeutic device response strategy; Post-processing the therapeutic device response strategy to obtain a therapeutic device response instruction, further converting the therapeutic device response instruction into a voice output to generate a final therapeutic device response information; The dialogue management model is constructed based on the deep reinforcement learning A3C algorithm, and a dialogue data set containing multiple dialogue scenarios is constructed as training data; wherein each dialogue scenario contains a user voice input sequence, a therapeutic device response sequence and a corresponding reward value; the user voice input sequence records each round of voice input information of the user during the dialogue process, and the therapeutic device response sequence records each round of response information given by the therapeutic device to the user input, and the reward value is set according to the quality of the dialogue, user satisfaction and task completion.

2. The intelligent voice interaction method of an osteoporosis therapeutic device according to claim 1, characterized in that: The steps to train a dialogue management model include: Step 1: Initialize the dialogue management model based on asynchronous advantage actor-critic A3C, define the state space of the model as the context information of the current dialogue, which includes the historical voice information of the user input, the historical information of the therapeutic device response, and the current state of the dialogue; define the action space of the model as the optional set of therapeutic device responses, which includes all response types or response templates given by the therapeutic device; define the reward function of the model as the dialogue quality score, which is used to calculate and distribute rewards during the training process; Step 2: Use the constructed dialogue dataset to train the dialogue management model. During the training process, an asynchronous update method is used for exploration and utilization. By continuously interacting with the environment, samples of states, actions, rewards, and transitions to the next state are collected to optimize the model’s decision-making strategy. Step 3: During the training process, calculate the average reward value of the model in the simulated dialogue; when the average reward value reaches the preset level, the training is completed and a trained dialogue management model is obtained.

3. The intelligent voice interaction method of an osteoporosis therapeutic device according to claim 1, characterized in that: The method of extracting features from speech data to obtain a speech feature vector comprises: Collect voice data of the user interacting with the therapeutic device; Preprocessing the collected speech data to obtain a preprocessed speech data set; the preprocessing step includes noise reduction and frame processing; The preprocessed speech data set is input into a speech processing model, wherein the input of the speech processing model is speech data and the output is a speech feature vector.

4. The intelligent voice interaction method of an osteoporosis therapeutic device according to claim 3, characterized in that: The training steps of the speech processing model include: Select the convolutional recurrent neural network CRNN model as the basic architecture of the speech processing model; Acquire speech training data, wherein the speech training data includes speech samples and their corresponding speech recognition labels; The CRNN model is trained using speech training data, and the CTC loss function is used for optimization during the training process; When the recognition accuracy of the speech processing model reaches a preset threshold, the training is stopped to obtain a trained speech processing model.

5. The intelligent voice interaction method for osteoporosis therapeutic apparatus according to claim 1, characterized in that: The training steps of the intent recognition model include: Select the long short-term memory network LSTM model as the basic architecture of the intent recognition model; Acquire intent training data, wherein the intent training data includes a speech feature vector and a corresponding user intent label; The LSTM model is trained using the intent training data, and the cross entropy loss function is used for optimization during the training process; When the accuracy of the intent recognition model reaches a preset threshold, the training is stopped to obtain a trained intent recognition model.

6. The intelligent voice interaction method for osteoporosis therapeutic apparatus according to claim 2, characterized in that: In the A3C algorithm, the actor network is used to generate action probability distribution, and the critic network is used to estimate the state value function. During the training process, the actor network selects actions according to the current strategy, and the critic network evaluates the value of the action. The actor network and the critic network are jointly optimized through the policy gradient and value function approximation methods.

7. The intelligent voice interaction method for osteoporosis therapeutic apparatus according to claim 1, characterized in that: The user voice input sequence and the therapeutic device response sequence are converted into text form through speech recognition and text conversion technology for training and reasoning of the dialogue management model.

8. The intelligent voice interaction method for osteoporosis therapeutic apparatus according to claim 7, characterized in that: The step of converting the therapeutic device response instruction into voice output comprises: The end-to-end text-to-speech synthesis model Tacotron is selected as the basic model. Its input is the text form of the therapeutic device response instructions, and its output is the synthesized speech signal. The therapeutic device response instructions are input into the text-to-speech synthesis model to generate the final speech response information.

9. The intelligent voice interaction method for osteoporosis therapeutic apparatus according to claim 1, characterized in that: The step of post-processing the therapeutic device response strategy comprises: Perform grammatical verification on the therapy device response strategy; Conducting contextual consistency checks on therapy device response strategies.

10. An intelligent voice interaction system for osteoporosis therapeutic apparatus, characterized in that: The system comprises: A voice data acquisition module is used to acquire the user's voice interaction data; A feature extraction module, connected to the voice data acquisition module, for extracting features from the acquired voice data to obtain a voice feature vector; An intention recognition module, connected to the feature extraction module, for inputting the speech feature vector into the intention recognition model to obtain the user intention; A dialogue management module, connected to the intention recognition module, constructed based on the deep reinforcement learning A3C algorithm, and used to input the user's intention into the dialogue management model to obtain the therapeutic device response strategy, wherein the training data of the dialogue management model includes a dialogue data set of constructed multi-round dialogue scenarios, each dialogue scenario includes a user voice input sequence, a therapeutic device response sequence, and a corresponding reward value; A post-processing module, connected to the dialogue management module, for post-processing the therapeutic device response strategy to obtain a therapeutic device response instruction; The voice output module is connected to the post-processing module and is used to convert the therapeutic device response instruction into a voice output to generate the final therapeutic device response information.

Images