An AI companionship interactive system

Abstract

The present invention discloses an AI companionship interactive system, which belongs to the technical field of data processing systems, and includes an audio receiving module, an audio noise reduction module, a data processing module and an audio amplification module connected in sequence; the audio receiving module is used to receive the user audio of the target user; the data processing module includes an audio converter, an audio semantic recognition unit, an audio generation unit and a data processor; the audio converter is connected to the audio generation unit through the audio semantic recognition unit, the audio semantic recognition unit is used to generate user semantics according to user audio data, and the audio generation unit is used to generate reply audio according to user semantics; the audio converter, the audio semantic understanding unit and the audio generation unit are all connected to the data processor; the audio amplification module is used to amplify the reply audio and feed it back to the target user. It can effectively reduce environmental noise and improve the accuracy of user intention recognition and response rate.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing systems, and specifically relates to an AI companionship interaction system. Background Art

[0002] The AI ​​companionship interactive system is a system based on artificial intelligence technology, which aims to simulate human interaction and provide emotional support, entertainment, companionship or information services through natural language processing, speech recognition, sentiment analysis and other technologies. It can run on various platforms (such as mobile applications, smart speakers, robots, etc.) to provide users with various interactive services. This system usually combines multiple fields of artificial intelligence, such as natural language processing (NLP), speech recognition and generation, machine learning, emotional computing, etc., to simulate human behavior and emotions and form an interactive experience similar to interpersonal communication.

[0003] In modern society, AI companionship and interaction systems help alleviate loneliness, improve mental health, and enhance quality of life by providing personalized emotional support, voice interaction, and real-time feedback. Especially for the elderly, singles, and those living away from their families, AI systems can not only provide emotional companionship, but also provide daily affairs management, entertainment, and health advice based on user needs, greatly enriching people's life experience and social interaction.

[0004] However, the existing AI companion interaction system is easily affected by environmental noise, resulting in slow response or even frequent response failures during the interaction process. In addition, the audio recognition accuracy of the AI ​​companion interaction system will be affected during the interaction process, leading to misrecognition of the user's intentions, which seriously affects the user experience. Summary of the invention

[0005] In order to solve the problem in the prior art that the AI ​​companion interaction system is easily affected by environmental noise, resulting in slow response or even frequent response failures in the interaction process, and the audio recognition accuracy of the AI ​​companion interaction system is affected during the interaction process, resulting in misrecognition of the user's intentions, seriously affecting the user experience, the present invention provides an AI companion interaction system.

[0006] The present invention provides an AI companionship interactive system, the AI ​​companionship interactive system comprising a secondary connected audio receiving module, an audio noise reduction module, a data processing module and an audio amplification module;

[0007] The audio receiving module is used to receive the user audio of the target user;

[0008] The data processing module includes an audio converter, an audio semantic recognition unit, an audio generation unit and a data processor. The audio noise reduction module is used to reduce the noise of the user audio, and the audio converter is used to convert the user audio after noise reduction into user audio data that can be recognized by the data processor;

[0009] The audio converter is connected to the audio generation unit through the audio semantic recognition unit, the audio semantic recognition unit is used to generate user semantics according to the user audio data, and the audio generation unit is used to generate a reply audio according to the user semantics;

[0010] The audio converter, the audio semantic understanding unit and the audio generation unit are all connected to the data processor;

[0011] The audio amplification module is used to amplify the reply audio and feed it back to the target user.

[0012] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0013] In the present invention, the AI ​​companion interactive system integrates an interconnected audio receiving module, an audio noise reduction module, a data processing module and an audio amplification module, and uses the audio noise reduction module in combination with the Bayesian algorithm and the gradient ascent method to reduce the noise of the user audio. It can adaptively filter the noise reduction parameters according to the user audio for adaptive noise reduction, rather than a fixed form of noise reduction, which greatly increases the audio noise reduction effect and improves the system's recognition accuracy and response speed to user intentions. In addition, it includes an audio converter, an audio semantic recognition unit, and an audio generation unit, all of which are connected to the data processing module, which can automatically respond to users and perform high-quality sound quality output, improve user experience, and improve the probability of successful response and the accuracy of intention recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The preferred implementation modes will be described below in a clear and understandable manner with reference to the accompanying drawings to further illustrate the above-mentioned characteristics, technical features, advantages and implementation methods of the present invention.

[0015] Figure 1 It is a structural schematic diagram of an AI companionship interactive system provided by the present invention;

[0016] Figure 2 The present invention is a flowchart of an audio noise reduction method. DETAILED DESCRIPTION

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings and other implementation methods can be obtained based on these drawings without creative work.

[0018] In order to simplify the drawings, only the parts related to the invention are schematically shown in each figure, and they do not represent the actual structure of the product. In addition, in order to simplify the drawings and facilitate understanding, in some figures, only one of the parts with the same structure or function is schematically drawn or marked. In this article, "one" not only means "only one", but also means "more than one".

[0019] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0020] In this document, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection. It can be a mechanical connection or an electrical connection. It can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0021] In addition, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0022] In one embodiment, referring to Figure 1 , showing a structural schematic diagram of an AI companionship interaction system provided by the present invention.

[0023] The present invention provides an AI companionship interactive system, which includes an audio receiving module, an audio noise reduction module, a data processing module and an audio amplification module connected in sequence.

[0024] Among them, the audio receiving module is responsible for receiving the user's audio signal, usually capturing sound through a microphone or other sound sensor. It is the "perception" part of the system and directly interacts with the user through audio. The audio noise reduction module is used to process noise in the audio signal. Noise is an unwanted signal in the audio, which usually interferes with the user's voice clarity and the system's response effect. The data processing module is the brain of the system, responsible for processing the audio signal received from the audio receiving module and generating corresponding feedback based on the analysis results. The main task of the audio amplification module is to amplify the generated reply audio and feed it back to the user through an audio output device (such as a speaker). These modules collaborate to enable the AI ​​companion interaction system to receive, process and feedback the user's audio requests, and achieve effective interaction and service.

[0025] The audio receiving module is used to receive the user audio of the target user.

[0026] The data processing module includes an audio converter, an audio semantic recognition unit, an audio generation unit and a data processor. The audio noise reduction module is used to reduce the noise of the user audio by combining the Bayesian algorithm and the gradient ascent method. The audio converter is used to convert the noise-reduced user audio into user audio data that can be recognized by the data processor.

[0027] The audio converter is connected to the audio generation unit through an audio semantic recognition unit. The audio semantic recognition unit is used to generate user semantics based on user audio data, and the audio generation unit is used to generate reply audio based on the user semantics.

[0028] The audio converter, the audio semantic understanding unit and the audio generation unit are all connected to the data processor.

[0029] The audio amplification module is used to amplify the reply audio and feed it back to the target user.

[0030] It should be noted that the AI ​​companion interaction system includes four main modules: audio receiving module, audio noise reduction module, data processing module and audio amplification module. The audio receiving module is responsible for capturing the user's audio signal, usually through a microphone, and transmitting the signal to the system. The audio noise reduction module uses the Bayesian algorithm and the gradient ascent method to remove noise from the audio signal to ensure clear audio. The data processing module is the core of the system, which includes an audio converter, a semantic recognition unit and an audio generation unit to process the denoised audio, recognize user semantics, and generate response audio. The audio amplification module is responsible for amplifying the generated reply audio and feeding it back to the user through the speaker. The entire system works together through these modules to complete the reception, processing and response of user audio requests, providing a smooth voice interaction experience.

[0031] In a possible implementation, the device further includes: a charging module.

[0032] The charging module is specifically an A01SB141B1-276 USB charging module.

[0033] The charging module is connected to the data processor and is used to supply power to the data processor.

[0034] It should be noted that the charging module uses the A01SB141B1-276 USB charging module, which is connected to the data processor to provide it with the required power support. This module ensures that the data processor can continue to be powered during system operation, thereby maintaining the normal operation of the system.

[0035] In a possible implementation manner, the system further includes: a communication module.

[0036] The communication module includes a wifi component and a bluetooth component.

[0037] The communication module is integrated in the data processor and is used to establish a data transmission path between the wireless data and the data processor.

[0038] It is understood that the communication module is integrated in the data processor and supports Wi-Fi and Bluetooth functions. It is used to establish a wireless data transmission path so that the data processor can communicate with external devices through a wireless network to transmit data or receive instructions.

[0039] In a possible implementation manner, the audio receiving module is specifically inmp441, which is used to receive user audio.

[0040] Among them, INMP441 is a high-performance digital microphone that uses an I2S interface and can convert sound into digital signals. It has the characteristics of low power consumption and high signal-to-noise ratio, and is often used as an audio input module in smart devices. INMP441 can effectively capture sound signals and convert them into digital format for subsequent processing. It is used to receive the user's audio signal. Through this module, the system can accurately capture the user's voice input, convert it into a digital signal, and transmit it to the downstream audio noise reduction and data processing modules for further processing.

[0041] In one possible implementation, the data processor is specifically an ESP32 processor, which is used to provide computing power for the audio converter, the audio semantic understanding unit, and the audio generation unit.

[0042] Among them, ESP32 is a low-power single-chip microcontroller with integrated Wi-Fi and Bluetooth functions, which is widely used in Internet of Things (IoT) devices. It has powerful computing power, built-in dual-core processor, rich peripheral interfaces, and supports high-speed data transmission. ESP32 also has efficient energy management and a variety of development support, suitable for applications that require wireless communication and low power consumption. It provides computing support for audio converters, audio semantic understanding units, and audio generation units. Through its powerful computing power, the ESP32 processor collaboratively processes audio signals, performs semantic recognition, and generates audio responses to achieve smooth voice interaction.

[0043] In a possible implementation manner, the audio amplification module is specifically a max98357 module, which is used to amplify the reply audio.

[0044] Among them, MAX98357 is an efficient digital audio amplifier that supports I2S input and can convert digital audio signals into analog signals and amplify them. It has low power consumption, high signal-to-noise ratio and excellent sound quality. It is often used in audio playback devices and can drive speakers to output clear audio signals. It is responsible for amplifying the reply audio output by the audio generation unit. Through this module, the system can effectively convert digital audio signals into analog signals, amplify them to an appropriate volume, and finally feed them back to the user through the speaker to ensure that the audio response is clear and audible.

[0045] In a possible implementation, the user audio is denoised by combining the Bayesian algorithm and the gradient ascent method, specifically including:

[0046] S1: Extract the time domain features of the user audio.

[0047] Among them, the time domain features include zero crossing rate, audio peak amplitude, audio minimum amplitude and audio average amplitude.

[0048] Among them, time domain features refer to the characteristics extracted by directly analyzing the changes of audio signals in time. They reflect the strength of the signal, frequency changes, etc., and are usually used to describe the basic characteristics of audio signals. Time domain analysis does not involve frequency distribution, but focuses on the fluctuation of the signal over time. The zero crossing rate refers to the number of times the audio signal crosses the zero axis per unit time. It reflects the high-frequency components of the audio signal. When the noise is strong or the signal changes rapidly, the zero crossing rate is usually high. The zero crossing rate is often used in the analysis of speech and audio signals to help distinguish between noise and speech. The peak amplitude refers to the maximum and minimum deviation from the zero point in the audio signal. It is usually used to describe the strength or dynamic range of the signal. When the peak amplitude is large, it means that the signal is strong. A smaller peak amplitude usually means that the signal is weak or there is a silent part. The minimum amplitude refers to the lowest amplitude value in the audio signal, which is usually a negative number. It is used to describe the weakest part of the signal. In contrast to the peak amplitude, the minimum amplitude reflects the lowest intensity level of the audio signal. The average amplitude refers to the average amplitude of the audio signal over a certain period of time, indicating the overall strength of the signal. A higher average amplitude generally indicates a stronger signal, while a lower average amplitude indicates a weaker signal. These time domain features can reflect the basic properties of the audio signal, such as frequency changes, volume intensity, and the dynamic range of the signal. By extracting these features, the system can better understand the nature of the audio signal, thereby providing a basis for subsequent noise estimation and processing.

[0049] S2: Based on the time domain features, the noise reduction parameters are determined by combining the Bayesian algorithm and the gradient ascent method, where the noise reduction parameters include the Fourier transform length and the frame shift.

[0050] Among them, the Bayesian algorithm is a statistical inference method based on Bayes' theorem, which is mainly used to update the probability distribution of model parameters. Bayes' theorem describes how to update the belief about an event or parameter after observing data. The gradient ascent method is an optimization algorithm used to find the maximum value of a function. In contrast to the gradient descent method, the gradient ascent method updates the parameters along the gradient direction (that is, the direction in which the function value increases fastest) to achieve the optimization goal. The Fourier transform length refers to the size of the data block used when performing the Fourier transform. In audio processing, the Fourier transform is used to convert audio signals from the time domain to the frequency domain to facilitate the analysis of its frequency components. The frame shift refers to the offset between the start position of each frame and the start position of the previous frame in audio processing. The Bayesian algorithm helps us estimate the appropriate noise reduction parameters by providing a combination of prior information and observed data, while the gradient ascent method adjusts these parameters through the optimization process to gradually improve the effect of audio noise reduction.

[0051] Understandably, a longer Fourier transform can capture more frequency information and is suitable for analyzing low-frequency changes. A shorter Fourier transform can reflect signal changes more quickly and is suitable for analyzing high-frequency changes. It is a balance between being able to identify details and not losing the general trend. When processing audio, the entire audio is usually not analyzed at once, but divided into many small segments, each of which is called a "frame". The frame shift determines how much overlap there is between each frame, or in other words, the offset of the start position of each frame relative to the previous frame. If the frame shift is small, adjacent frames will have a large overlap, which can provide more continuous analysis and is suitable for analyzing fast-changing audio (such as fast pitch changes in speech). If the frame shift is large, there is less overlap between frames and the analysis speed is faster, but some details may be lost.

[0052] In a possible implementation, S2 specifically includes:

[0053] S201: Acquire historical user audio sample data, wherein the historical user audio sample data includes historical time domain features and noise reduction parameter labels corresponding to the time domain features.

[0054] The labels can be manually annotated. The goal is to infer the denoising parameters of unseen data from limited data.

[0055] In actual applications, the characteristics of noise change dynamically, especially in environmental noise or real-time calls, where the frequency, intensity, and even pattern of noise may change rapidly. Therefore, the noise prior probability distribution constructed based on historical data may not be able to perfectly adapt to the noise changes at all times.

[0056] S202: Establish a Bayesian formula based on historical user audio sample data to determine the prior probability of noise reduction parameters:

[0057]

[0058] Among them, p(θ|X) represents the posterior probability, that is, the probability of the noise reduction parameter θ given the historical time domain feature X, p(X|θ) represents the likelihood function, that is, the probability of observing X under given θ, p(θ) represents the prior probability of θ, that is, the probability that the noise reduction parameter is θ when the data X is not observed, and p(X) represents the marginal likelihood used to ensure that the posterior probability of all noise reduction parameters is 1.

[0059] Among them, the specific meaning of marginal likelihood is to integrate all possible θ to obtain the total probability of observed data X.

[0060] S203: Obtain time domain features, and construct an objective function for optimizing the noise reduction parameters according to the prior probability of the noise reduction parameters:

[0061] log(P(D|θ))+log(P(θ))

[0062] Where log represents the logarithmic function, D represents the time domain feature, and (P(D|θ)) represents the probability of observing D under a given denoising parameter θ.

[0063] It should be noted that the objective function is equal to the logarithmic form of the part of the posterior probability related to θ. The reason for taking the logarithm is to convert multiplication into addition, simplify the calculation, and will not change the position of the extreme point.

[0064] S204: With the goal of maximizing the posterior probability of the denoising parameter for the time domain feature, the denoising parameter is updated by the gradient ascent method until the updated denoising parameter is stabilized in a preset range, thereby obtaining the denoising parameter for denoising the time domain feature:

[0065]

[0066] Among them, θ t+1 They represent the denoising parameters obtained at the t-th and t+1-th iterations, α represents the learning rate, represents the partial derivative of the objective function log(P(D|θ))+log(P(θ)) with respect to θ.

[0067] It should be noted that those skilled in the art can set the size of the preset interval according to actual needs, and the present invention is not limited here. It is understandable that in the process of gradient ascent solution, the noise reduction parameter will stably converge within a certain fluctuation range, that is, within the preset interval. At this time, the iteration can be stopped and the noise reduction parameter in the iterative process is used as the optimized noise reduction parameter.

[0068] Optionally, Among them, μ represents the mean of the noise reduction parameter, σ represents the variance of the noise reduction parameter, π represents pi, and exp represents the natural exponential function.

[0069]

[0070] Specifically, the noise reduction parameters are optimized by combining Bayesian inference and gradient ascent method. First, through historical audio data, the system obtains the time domain features of the audio (such as zero crossing rate, peak amplitude, etc.) and the noise reduction parameter labels corresponding to these features, which may be manually labeled. Through these limited data, the system infers the noise reduction parameters under unseen audio features. Then, based on these historical data, the Bayesian formula is established to calculate the prior probability of the noise reduction parameters, that is, the distribution of the noise reduction parameters given the historical data. Then, using these time domain features and prior probabilities, an objective function for optimizing the noise reduction parameters is constructed, the purpose is to maximize the posterior probability of the noise reduction parameters. The objective function is optimized by the gradient ascent method, and the noise reduction parameters are gradually adjusted until the parameters are stable within the preset range, thereby obtaining the optimal noise reduction parameters. This process combines Bayesian inference with optimization algorithms to ensure that the noise reduction parameters can be effectively adjusted in a dynamically changing noise environment to obtain the best noise reduction effect.

[0071] S3: Divide the user audio into frames based on the noise reduction parameters to obtain the original spectrum of each frame.

[0072] Specifically, the audio signal itself is continuous, but when performing frequency domain analysis, it is usually necessary to divide it into multiple short segments (frames). Each frame represents a small time window of the audio signal. The size of the frame is usually determined by the Fourier transform length (a noise reduction parameter). Longer frames can capture lower frequency components, while shorter frames are suitable for capturing rapidly changing high frequency parts. The frame shift (another noise reduction parameter) determines the overlap between each frame. If the frame shift is small, adjacent frames will have more overlap, which helps to provide a more continuous analysis and is suitable for rapidly changing signals (such as speech). The audio signal is framed according to the noise reduction parameters (such as Fourier transform length and frame shift), and then each frame is Fourier transformed to obtain the original spectrum of each frame, which provides a basis for subsequent noise estimation and spectrum processing. This process converts the audio signal in the time domain into frequency domain information, which helps to effectively identify and process the noise components in the audio.

[0073] S4: extract the silent spectrum from the original spectrum of each frame.

[0074] The silent spectrum refers to the spectrum component of the silent or background noise part in the audio signal. It can be extracted through librosa, scipy or pydub.

[0075] S5: Estimate the noise spectrum of the original spectrum corresponding to the silence spectrum based on the silence spectrum.

[0076] Specifically, there should be no audio signal in the silent spectrum, and the signals that appear are all caused, and the audio signals that appear in the silent spectrum are all used as the noise spectrum of the original spectrum.

[0077] S6: removing the noise spectrum in the original spectrum of each frame to obtain multiple frames of denoised spectrum.

[0078] S7: Reconstruct the denoised spectrum and output the denoised user audio.

[0079] In actual application, first, the audio signal is processed in frames, and the spectrum of each frame is obtained through Fourier transform. Next, by extracting the silence spectrum of each frame, we can identify the silence part or background noise in the audio. Then, the silence spectrum is used to estimate the spectrum of the noise in the audio, assuming that the signal part that appears in the silence segment is the noise spectrum. Next, by removing the noise spectrum from the original spectrum, the denoised spectrum is obtained. Finally, by reconstructing the denoised spectrum, the final clear audio is output. Through this series of steps, the system can effectively reduce the noise component and improve the audio quality.

[0080] In a possible implementation, user semantics is generated according to user audio data, specifically:

[0081] Input user audio data into the conversation model and output user semantics.

[0082] Optionally, the dialogue model is specifically the Tongyi Qianwen dialogue model or the Wenxin Yiyan dialogue model.

[0083] It should be noted that the user audio data is first analyzed by the audio processing module, and then input into the dialogue model, such as Tongyi Qianwen or Wenxin Yiyan, to generate the user's semantics. These dialogue models convert audio into structured semantic information through natural language processing technology, so that the system can understand the user's intentions and needs and make appropriate responses. Through this process, the AI ​​system can provide a smarter and more natural voice interaction experience.

[0084] In a possible implementation, generating a reply audio according to user semantics specifically includes:

[0085] Input user semantics into the conversation model to generate response text.

[0086] The reply text is input into the speech synthesis model and the reply audio is output.

[0087] Optionally, the speech synthesis model is specifically a WaveNet model, a Tacotron model or a FastSpeech model.

[0088] It should be noted that after the user's semantics is input into a large dialogue model (such as Tongyi Qianwen or Wenxin Yiyan), a reply text is generated. Subsequently, the reply text is passed to a large speech synthesis model (such as WaveNet, Tacotron, or FastSpeech), which converts the text into a natural voice signal to generate a reply audio. This process enables the system to generate appropriate responses based on the user's semantic understanding and feedback to the user in the form of voice, providing a smooth voice interaction experience.

[0089] In actual application, the AI ​​companion interactive system realizes intelligent voice interaction between users and the system through four main modules (audio receiving module, audio noise reduction module, data processing module and audio amplification module). First, the audio receiving module captures the user's audio signal through INMP441 and transmits it to the audio noise reduction module, which uses the Bayesian algorithm and gradient ascent method to suppress noise and improve audio quality. Then, the denoised audio is semantically analyzed and processed by the data processing module, and the user semantics are generated using a large dialogue model (such as Tongyi Qianwen or Wenxin Yiyan), and a suitable reply text is generated. Finally, the reply text is converted into speech through a large speech synthesis model (such as WaveNet, Tacotron or FastSpeech), output to the audio amplification module, and fed back to the user through the speaker amplification to achieve a natural and smooth voice interaction experience. At the same time, the system also integrates a charging module and a communication module to ensure the continuous power supply and wireless data transmission function of the device. The system can provide personalized voice interaction services to adapt to changing environmental noise and user needs.

[0090] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0091] In the present invention, the AI ​​companion interactive system integrates an interconnected audio receiving module, an audio noise reduction module, a data processing module and an audio amplification module, and uses the audio noise reduction module in combination with the Bayesian algorithm and the gradient ascent method to reduce the noise of the user audio. It can adaptively filter the noise reduction parameters according to the user audio for adaptive noise reduction, rather than a fixed form of noise reduction, which greatly increases the audio noise reduction effect and improves the system's recognition accuracy and response speed to user intentions. In addition, it includes an audio converter, an audio semantic recognition unit, and an audio generation unit, all of which are connected to the data processing module, which can automatically respond to users and perform high-quality sound quality output, improve user experience, and improve the probability of successful response and the accuracy of intention recognition.

[0092] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An AI companionship interactive system, characterized in that: The AI ​​companionship interactive system includes an audio receiving module, an audio noise reduction module, a data processing module and an audio amplification module connected in sequence; The audio receiving module is used to receive the user audio of the target user; The data processing module includes an audio converter, an audio semantic recognition unit, an audio generation unit and a data processor, the audio noise reduction module is used to combine the Bayesian algorithm and the gradient ascent method to reduce the noise of the user audio, and the audio converter is used to convert the user audio after noise reduction into user audio data that can be recognized by the data processor; The audio converter is connected to the audio generation unit via the audio semantic recognition unit, the audio semantic recognition unit is used to generate user semantics according to the user audio data, and the audio generation unit is used to generate a reply audio according to the user semantics; The audio converter, the audio semantic understanding unit and the audio generation unit are all connected to the data processor; The audio amplification module is used to amplify the reply audio and feed it back to the target user.

2. The AI ​​companionship interactive system according to claim 1, characterized in that: Also includes: Charging module; The charging module is specifically an A01SB141B1-276 USB charging module; The charging module is connected to the data processor and is used to supply power to the data processor.

3. The AI ​​companionship interactive system according to claim 1, characterized in that: Also includes: Communication module; The communication module includes a wifi component and a bluetooth component; The communication module is integrated with the data processor and is used to establish a data transmission path between wireless data and the data processor.

4. The AI ​​companionship interactive system according to claim 1, characterized in that: The audio receiving module is specifically inmp441, which is used to receive the user audio.

5. The AI ​​companionship interactive system according to claim 2, characterized in that: The data processor is specifically an ESP32 processor, which is used to provide computing power for the audio converter, the audio semantic understanding unit and the audio generation unit.

6. The AI ​​companionship interactive system according to claim 5, characterized in that: The audio amplification module is specifically a max98357 module, which is used to amplify the reply audio.

7. The AI ​​companionship interactive system according to claim 1, characterized in that: The combining of the Bayesian algorithm and the gradient ascent method to reduce noise on the user audio specifically includes: S1: extracting time domain features of the user audio, wherein the time domain features include zero crossing rate, audio peak amplitude, audio minimum amplitude and audio average amplitude; S2: Determine noise reduction parameters based on the time domain features in combination with the Bayesian algorithm and the gradient ascent method, wherein the noise reduction parameters include Fourier transform length and frame shift; S3: dividing the user audio into frames based on the noise reduction parameter to obtain an original spectrum of each frame; S4: extract the silent spectrum from the original spectrum of each frame; S5: estimating a noise spectrum of an original spectrum corresponding to the silence spectrum based on the silence spectrum; S6: removing the noise spectrum in the original spectrum of each frame to obtain multiple frames of denoised spectrum; S7: Reconstruct the denoised spectrum and output the denoised user audio.

8. The AI ​​companionship interactive system according to claim 7, characterized in that: The S2 specifically includes: S201: Acquire historical user audio sample data, wherein the historical user audio sample data includes historical time domain features and noise reduction parameter labels corresponding to the time domain features; S202: Establishing a Bayesian formula based on the historical user audio sample data to determine the prior probability of the noise reduction parameter: Among them, p(θ|X) represents the posterior probability, that is, the probability of the noise reduction parameter θ given the historical time domain feature X, p(X|θ) represents the likelihood function, that is, the probability of observing X under a given θ, p(θ) represents the prior probability of θ, that is, the probability that the noise reduction parameter is θ when the data X is not observed, and p(X) represents the marginal likelihood used to ensure that the posterior probability of all noise reduction parameters is 1; S203: Acquire the time domain features, and construct an objective function for optimizing the noise reduction parameters according to the prior probability of the noise reduction parameters: log(P(D|θ))+log(P(θ)) Where log represents the logarithmic function, D represents the time domain feature, and (P(D|θ)) represents the probability of observing D under a given noise reduction parameter θ; S204: With the goal of maximizing the posterior probability of the denoising parameter under the time domain feature, the denoising parameter is updated by a gradient ascent method until the updated denoising parameter is stabilized in a preset interval, thereby obtaining the denoising parameter for denoising the time domain feature: i t+1 =θ t +a·▽ θ (log(P(D|θ))+log(P(θ))) Among them, θ t+1 denote the denoising parameters obtained at the t-th and t+1-th iterations, α denotes the learning rate, ▽ θ represents the partial derivative of the objective function log(P(D|θ))+log(P(θ)) with respect to θ.

9. The AI ​​companionship interactive system according to claim 1, characterized in that: Generating user semantics according to the user audio data is specifically: The user audio data is input into the dialogue model, and the user semantics is output.

10. The AI ​​companionship interactive system according to claim 9, characterized in that: The step of generating a reply audio according to the user semantics specifically includes: Inputting the user semantics into the dialogue macro model to generate a reply text; The reply text is input into a speech synthesis model, and the reply audio is output.

Images