Intelligent service robot system integrated with voice recognition

Abstract

The invention relates to the technical field of intelligent robot auditory perception and voice interaction, in particular to an intelligent service robot system integrated with voice recognition. The system comprises a multi-dimensional data acquisition module, a dynamic simulation processing module and a signal coupling judgment module. The system constructs standard semantic features by synchronously acquiring acoustic signals and kinematic telemetry data; the core of the method is that acoustic channel distortion simulation is carried out on standard features in combination with a motion state, theoretical distortion features are generated, and the similarity between the theoretical distortion features and real signals is calculated; if the similarity exceeds a threshold value, recovering the instruction, otherwise, performing noise filtering; according to the method, the transformation from passive noise reduction to active simulation verification based on a physical model is realized, the problem of non-stationary noise interference generated by robot motion is solved, and the interaction robustness under extreme working conditions is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of auditory perception and voice interaction technology for intelligent robots, specifically to an intelligent service robot system integrating voice recognition. Background Technology

[0002] In human-computer interaction applications of intelligent service robots, the accuracy of speech recognition systems is limited by complex acoustic environments. In particular, when the robot is in navigation or motion, it is necessary to process acoustic signal acquisition and command parsing simultaneously. Existing technologies generally adopt passive noise reduction algorithms or statistical general acoustic models, focusing on suppressing steady-state background noise through spectral subtraction or fixed beamforming. However, this approach has significant limitations during the dynamic operation of robots. Due to the non-stationary mechanical noise generated by the robot's own movement speed and motor operation, as well as the spectral distortion caused by the Doppler effect, traditional methods lack active physical modeling based on kinematic parameters. They cannot effectively distinguish between the Doppler frequency shift caused by the robot's own motion, the motor masking effect, and the true spectral characteristics of the user's speech. When faced with multipath reverberation and rapid speed change scenarios, existing models are unable to correct time-varying phase drift. They are prone to misjudging effective speech signals affected by motion as environmental noise and filtering them out, or failing to lock weak command signals from high-intensity motor noise, leading to interaction response failures or increased misrecognition rates. Therefore, how to accurately decouple and verify the motion noise of the user's speech spectrum distortion under dynamic motion conditions in order to improve the robustness and accuracy of speech recognition has become an urgent technical problem to be solved. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides an intelligent service robot system integrating speech recognition. Specifically, the technical solution of this invention includes: Multidimensional data acquisition module, dynamic simulation processing module, and signal coupling decision module; The multidimensional data acquisition module is used to acquire the original acoustic signal data of the robot's current environment and the kinematic telemetry data of the robot, and send the original acoustic signal data and the kinematic telemetry data to the dynamic simulation processing module. The dynamic simulation processing module is used to construct standard semantic reference features based on the current interaction context, and combine the kinematic telemetry data to perform acoustic channel distortion simulation on the standard semantic reference features in order to generate theoretical distortion features. The signal coupling decision module is used to extract real signal features from the original acoustic signal data and calculate the signal similarity between the real signal features and the theoretical distortion features; The configuration is as follows: if the signal similarity is greater than a preset coupling threshold, the original acoustic signal data is determined to contain a valid voice command, and the original acoustic signal data is restored based on the theoretical distortion characteristics to output the voice command; otherwise, the original acoustic signal data is determined to be environmental noise and a filtering operation is performed.

[0004] Preferably, the dynamic simulation processing module is further used for: Based on the current interaction context, candidate keywords are extracted; Generate standard acoustic feature vectors for the candidate keywords in a preset anechoic chamber environment, which serve as the standard semantic reference features.

[0005] Preferably, the dynamic simulation processing module is further used for: Based on the kinematic telemetry data, the motor operating state parameters and movement speed vector of the robot are analyzed. Based on the motor operating state parameters, a preset motor noise harmonic library is queried to synthesize the self-noise spectrum; Based on the moving velocity vector and the preset environmental prior map, the dynamic transfer function from the sound source to the microphone is calculated; The standard semantic reference features are convolved or processed using the dynamic transfer function, and the self-noise spectrum is superimposed to generate the theoretical distortion features.

[0006] Preferably, the signal coupling decision module is further used for: Obtain a preset background noise reference model; The difference features between the original acoustic signal data and the background noise reference model are calculated as the real signal features; Calculate the difference features between the theoretical distortion features and the standard semantic reference features, and use them as theoretical residual features; The cross-correlation coefficient or structural similarity index between the actual signal features and the theoretical residual features on the time-frequency spectrum is calculated as the signal similarity.

[0007] Preferably, the signal coupling decision module is further used for: If it is determined that the original acoustic signal data contains valid speech commands, the corresponding inverse filter parameters are calculated based on the theoretical distortion characteristics. The inverse filtering parameters are applied to the original acoustic signal data to restore a clean speech signal.

[0008] Preferably, the multidimensional data acquisition module includes: a microphone array and a motion parameter detection unit; The microphone array is used to acquire the raw acoustic signal data, which includes time-domain waveform data and frequency-domain feature data. The motion parameter detection unit is used to collect the kinematic telemetry data, which includes chassis speed, acceleration, head servo angle, and motor PWM duty cycle.

[0009] Preferably, the system further includes: a control execution module; The control execution module is used to receive the voice command output by the signal coupling decision module, and control the robot to perform corresponding service actions according to the voice command.

[0010] Preferably, the dynamic simulation processing module is further used for: When the robot's movement velocity vector changes, the dynamic transfer function is updated in real time to adjust the generation parameters of the theoretical distortion feature.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This system synchronously acquires acoustic signals and robot kinematic telemetry data through a multi-dimensional data acquisition module, transforming traditional passive noise reduction into active simulation verification based on a physical model. The system can actively construct theoretical distortion characteristics that include the Doppler effect and mechanical noise by utilizing parameters such as the robot's own moving speed and motor operating status. This mechanism effectively solves the problem that existing technologies cannot distinguish between non-stationary noise generated by the robot's own motion and the spectral characteristics of the user's voice when the robot is navigating or moving at high speed. This enables the robot to accurately lock onto weak command signals even under extreme conditions such as full-speed movement or high-intensity motor operation, significantly improving the robustness of voice interaction. 2. This system, through the collaborative work of the dynamic simulation processing module and the signal coupling decision module, introduces standard acoustic characteristics based on the reverberationless anechoic chamber environment as a purity benchmark, and combines environmental prior maps with real-time motion parameters to generate a theoretical prediction model. By calculating the similarity between real signal characteristics and theoretical distortion characteristics in the time-frequency spectrum, the system performs rigorous topological verification. This method avoids the problem of traditional statistical models misidentifying effective speech affected by motion as environmental noise and filtering it out due to phase drift or spectral distortion, effectively reducing the false recognition rate and ensuring the sensitive capture of effective speech commands in complex sound field environments. 3. After determining that the signal contains valid speech commands, the system calculates inverse filtering parameters using the generated theoretical distortion features and applies them to the original acoustic signal. Since the inverse filtering parameters are calculated based on the precise motion state and sound field propagation model at the current moment, they can specifically cancel out the specific frequency shift and masking effects caused by motion. This inverse restoration technique based on deterministic physical processes, compared with blind source separation or general filtering algorithms, can better remove non-stationary mechanical noise and reverberation interference, restore a high-purity speech signal, and ensure the accuracy of subsequent semantic understanding. 4. This system integrates a real-time monitoring mechanism for changes in the moving speed vector. Once the robot accelerates, stops abruptly, or turns, the dynamic transfer function is updated in real time and the generation parameters of the theoretical distortion characteristics are adjusted. Combined with the motor noise harmonic library and the instantaneous geometric path calculation method, the system eliminates non-physical phase drift caused by time accumulation, ensuring millisecond-level synchronization between the simulation model and physical reality. This enables the robot to adapt to rapidly changing dynamic sound fields and maintain stable tracking and recognition of command signals during non-stationary motion, achieving zero-delay closed-loop control from perception to execution. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1: Please see Figure 1 The intelligent service robot system integrating speech recognition includes: a multi-dimensional data acquisition module, a dynamic simulation processing module, and a signal coupling decision module; Among them, the multi-dimensional data acquisition module is used to acquire the original acoustic signal data of the robot's current environment and the robot's kinematic telemetry data, and send the original acoustic signal data and kinematic telemetry data to the dynamic simulation processing module; The dynamic simulation processing module is used to construct standard semantic reference features based on the current interaction context, and combine kinematic telemetry data to perform acoustic channel distortion simulation on the standard semantic reference features in order to generate theoretical distortion features; The signal coupling decision module is used to extract real-world signal features from the raw acoustic signal data and calculate the signal similarity between the real-world signal features and the theoretical distortion features; The configuration is as follows: if the signal similarity is greater than the preset coupling threshold, the original acoustic signal data is determined to contain valid speech commands, and the original acoustic signal data is restored based on theoretical distortion characteristics to output speech commands; otherwise, the original acoustic signal data is determined to be environmental noise and a filtering operation is performed.

[0015] This embodiment details the specific architecture and operational logic of the aforementioned system. This system aims to solve the technical challenge of existing service robots being unable to effectively distinguish between the Doppler effect generated by the robot's own movement and mechanical noise, as well as the true spectral distortion of the user's speech during dynamic motion. The system executes the perception front-end task through a multi-dimensional data acquisition module. This module collects noisy frequency streams from the robot's microphone array in real time and performs preprocessing to output raw acoustic signal data containing both time-domain waveforms and frequency-domain features. Simultaneously, it reads the robot's motion parameters from the robot chassis control bus as kinematic telemetry data. These two types of data are then stamped with a unified timestamp and sent to the next level. The dynamic simulation processing module, as the core inference engine, employs synthesis analysis logic based on the current interaction context. Extracting high-probability answers to construct standard semantic reference features And combined with kinematic telemetry data Acoustic channel distortion simulation was performed on it to generate theoretical distortion characteristics. This feature characterizes the spectrum of the damaged signal that the microphone should receive after the user speaks standard speech, following the Doppler shift caused by the current motion speed and the noise masking generated by the motor speed. The signal coupling decision module performs topological verification of true and false signals from the raw acoustic signal data. Extracting real-world signal features And calculate its distortion characteristics compared with the theoretical distortion characteristics. Signal similarity between ; in response Greater than the preset coupling threshold This threshold is not arbitrarily set, but rather determined during the system's offline calibration phase by collecting pure noise samples and command-bearing samples of the robot running at speeds between 0 and 1.5 m / s. The threshold corresponding to a false recognition rate (FAR) of <0.5% is selected based on ROC curve analysis, for example, set to 0.82. The system then determines that the signal contains valid voice commands and performs inverse reconstruction based on distortion parameters. Less than or equal to The system identifies it as environmental noise and filters it out. This embodiment introduces kinematic telemetry data for active simulation, transforming traditional passive noise reduction into expectation verification based on a physical model. This enables the robot to accurately lock onto user commands even under extreme conditions such as full-speed movement or motor noise, avoiding the problem of cutting out effective speech due to misjudgment of motion noise.

[0016] Example 2: The dynamic simulation processing module is also used for: Extract candidate keywords based on the current interaction context; Generate standard acoustic feature vectors for candidate keywords in a preset anechoic chamber environment, which serve as standard semantic reference features.

[0017] This embodiment generates standard semantic reference features using the dynamic simulation processing module in claim 1. Further details of the process: The system maintains a dynamic dialogue state machine based on the current interaction context. The system extracts candidate keywords; for example, when the robot is in a navigation confirmation state, the system locks in high-probability words such as "yes," "no," and "depart." For each extracted candidate keyword, the system calls a preset anechoic chamber acoustic database, which stores the standard acoustic feature vectors of each keyword in an environment of absolute stillness and no reverberation, i.e., T60 < 0.05s. The system directly maps these feature vectors to standard semantic reference features. This serves as a clean baseline for subsequent simulations; This embodiment establishes the pure ideal state of the signal by using the characteristics of a reverberant anechoic chamber environment as a benchmark, providing an accurate zero-point reference for subsequent superposition of motion distortion, thereby significantly improving the physical realism of the simulation results. To ensure that those skilled in the art can write code to generate and invoke this feature, and to eliminate dimensional ambiguity in subsequent mathematical operations, this embodiment explicitly defines the specific parameters of the standard acoustic feature vector: the data type is a single-precision complex tensor that preserves phase information, such as complex64; the extraction algorithm uses a short-time Fourier transform (STFT), and the sampling rate is configured... The window function is a Hanning window, and the frame length is... That is, 400 points, frame shift That is, 160 points, FFT points 512; the corresponding data dimension is ,in, The time frame number is 257, and the frequency point number is 257. This definition ensures that the feature vector can directly participate in the channel convolution operation based on the complex field in Example 3, and the difference feature calculation based on amplitude in Example 4. The dynamic simulation processing module is also used for: Based on kinematic telemetry data, the robot's motor operating state parameters and movement speed vector are analyzed. Based on the motor's operating state parameters, a preset motor noise harmonic library is queried to synthesize the self-noise spectrum; Based on the moving velocity vector and a pre-set environmental prior map, the dynamic transfer function from the sound source to the microphone is calculated; The standard semantic reference features are convolved or processed in the frequency domain using a dynamic transfer function, and then superimposed with a self-noise spectrum to generate theoretical distortion features.

[0018] This embodiment describes the generation theory distortion feature in claim 2. The process is further specified, detailing the reconstruction of the physical field and signal processing; the system uses kinematic telemetry data... The key motion variables are extracted from the data; to strictly conform to the laws of physics and eliminate ambiguity, the system performs the following kinematic chain calculation: reads the chassis linear velocity vector. and head servo angle The instantaneous velocity vector in the microphone array coordinate system is calculated using the rotation matrix. ; Analyze the current time frame Motor operating status parameters This refers to the PWM duty cycle; to simulate the robot's own mechanical noise, the system constructs a time-varying self-noise model based on harmonic superposition; to ensure the physical integrity of the simulation signal in the complex domain, the system calculates the amplitude spectrum of the current frame, assigns an independent random phase to the frame, and explicitly defines the amplitude values ​​in the preset harmonic library. The system has been calibrated offline using a standard sound pressure meter, completing the dimensional mapping from physical sound pressure (Pa) to the system's digital quantization value (DigitalLevel). This ensures that the synthesized noise and the actual acquired signal are within the same dynamic range and can change over time. The formula is dynamically updated based on changes in the motor's state, as follows: ; in, The source is computationally generated, and the physical meaning is within a time frame. The complex synthesis self-noise spectrum with lower-band phase; The source is a pseudo-random number generator, for each time frame. Independently generated, subject to The uniform distribution is used to simulate the random phase characteristics of incoherent noise; The source is the motion parameter detection unit at time... The sampled value, in physical terms, represents the motor PWM duty cycle within the normalized range of 0-1; The source is a pre-defined motor noise harmonic library, i.e., linear amplitude; The source is a preset harmonic cutoff order, which is 10 in this embodiment; The physical meaning is a discretized impact function; in the code implementation, it is represented by the amplitude... Accumulate to the FFT frequency point of the corresponding frequency index; The source is based on the current time. The broadband wind noise component generated by the moving speed; to ensure the consistency of the formula dimensions, this embodiment adopts a normalized spectrum model: ; Among them, cutoff frequency , ; This is clearly stated The physical meaning of is the aeroacoustic coupling coefficient, and its dimension is strictly defined as . Used to offset the squared velocity term Dimensions This ensures that the calculation result is a dimensionless amplitude value; this parameter characterizes the conversion efficiency of kinetic energy per unit velocity squared into microphone diaphragm vibration energy; this value is not a universal constant, but an empirical value obtained by calibrating a specific microphone structure in a wind tunnel, used to map the hydrodynamic quantities of the physical world to the amplitude domain of the digital signal processing system. Based on the calculated velocity vector and the pre-set environmental prior map, the dynamic transfer function from the sound source to the microphone is calculated. Considering that traditional global Doppler models cannot accurately describe the physical fact that different reflection paths within a reverberation chamber have different incident angles and relative velocities, this embodiment employs a multipath differential Doppler model; the system iterates through each virtual sound source generated by the mirror source method. , For direct sound, For reflected sound, calculate their independent dynamic phase shifts; to correct the original formula which used a linear velocity approximation. The fundamental error caused by phase drift with absolute time is addressed in this embodiment using the instantaneous geometric path calculation method: Geometric solution: For each mirror source According to its spatial coordinates and the robot microphone array in the current time frame precise coordinates Calculate instantaneous sound path Transfer function synthesis: The phase delay is directly calculated using the instantaneous sound path, and the formula is modified as follows: ; in, : Number of mirror sources to be truncated, for example, 1000; :No. The number of times the path experiences wall reflections; The source is a preset environmental prior map attribute. The environmental prior map is constructed as a vector database containing environmental geometric coordinates and corresponding surface material semantic labels. The physical meaning is the dimensionless average sound absorption coefficient of the wall. In this embodiment, the system indexes the nearest wall material label in the vector map according to the microphone coordinates and looks up the value in the table. For example, 0.05 is taken for concrete material and 0.12 is taken for glass material to ensure that the energy attenuation model conforms to the real physical environment. The source is a physical constant, and its value is... That is, the speed of sound under standard atmospheric pressure, which serves as the benchmark for time delay and wavelength conversion; : characterizes the first The path at time The instantaneous physical path length; this term is directly updated by geometric relations, replacing the original linear velocity integral term, thereby eliminating the non-physical phase drift caused by time accumulation and accurately reflecting the instantaneous frequency response change caused by the Doppler effect; The system generates simulation features by performing frequency domain response processing and noise superposition on an ideal reference. The formula is as follows: ; This embodiment corrects the physical defects of the single Doppler factor by introducing a differential Doppler model based on instantaneous geometric paths, thus ensuring the mathematical accuracy of theoretical distortion characteristics in complex sound fields.

[0019] Example 3: The signal coupling decision module is also used for: Obtain a preset background noise reference model; Calculate the difference features between the original acoustic signal data and the background noise reference model, and use them as features of the real signal; The difference between the theoretical distortion features and the standard semantic reference features is calculated and used as the theoretical residual features; The cross-correlation coefficient or structural similarity index between the real signal characteristics and the theoretical residual characteristics on the time-frequency spectrum is calculated as the signal similarity.

[0020] This embodiment further specifies the dual-track differential extraction and similarity calculation logic of the signal coupling decision module in claim 1; in order to eliminate the interference of constant background noise, the system obtains a preset background noise reference model. The storage format is amplitude spectrum; during operation, the system calculates the short-time energy of the current acoustic signal frame in real time, and updates it if it falls below a threshold. The specific update logic employs a recursive averaging method to ensure smooth adaptation of the noise model. ; in, The preset forgetting factor, set to 0.95, ensures that the model learns only steady-state background noise while ignoring transient speech bursts; the system calculates... Compared with the original acoustic signal data The difference; to avoid phase errors caused by direct subtraction of complex numbers, amplitude spectrum subtraction is explicitly performed here to extract the characteristics of the real signal. The formula is as follows: ; in, The short-time Fourier transform amplitude spectrum of the original signal. It is a half-wave rectification operator used to remove non-physical negative values ​​generated by over-subtraction; Simultaneously, the system calculates theoretical distortion characteristics. With standard semantic reference features The difference between them; similarly, in order to compare the degree of distortion under the same dimensions, the difference of the amplitude spectrum is calculated here to obtain the theoretical residual characteristics. : ; This feature characterizes the incremental texture that should theoretically be produced in the frequency spectrum due to the Doppler effect and motor noise; To verify Does the included distortion conform to The system calculates the cross-correlation coefficient between the two signals on the time-frequency spectrum as the signal similarity. It should be noted that, due to It is a continuous time flow characteristic. It is a fixed-length template. Before executing the following formulas, the system uses a sliding window matching strategy: exist Slide along the timeline; to eliminate ambiguity in the summation range, this embodiment explicitly defines the calculation logic within the local matching window, assuming... The time frame length is Currently swiping to The starting frame index is The formula for single alignment is as follows. To prevent numerical calculation crashes caused by a denominator of zero, a regularization term is introduced. : ; in, The source is a calculated value; its physical meaning is at time [time]. The local signal similarity, with values ​​ranging from [0,1]; : Clearly defined as a double accumulation operation: the inner layer performs accumulation on all frequency points from 0Hz to the Nyquist frequency. Summation; outer layer local time index Summation, where, Corresponding to theoretical residual characteristics The Frame and real signal characteristics The first in the current window Frame; this definition ensures that similarity calculations are performed only on the currently covered local time-frequency blocks, not the entire sequence; The source is a preset constant, for example ; In response to A value close to 1 indicates the distribution of noise and distortion in the real-world signal. The physical field distortion predicted by theory The signal was highly consistent, thus confirming that it was valid speech contaminated by robot movement. This embodiment eliminates logical ambiguity in complex number operations by strictly defining amplitude spectrum subtraction and rectification operations, thus ensuring the programmability of the algorithm. The signal coupling decision module is also used for: If the original acoustic signal data is determined to contain valid speech commands, the corresponding inverse filter parameters are calculated based on theoretical distortion characteristics. The inverse filtering parameters are applied to the original acoustic signal data to restore the clean speech signal.

[0021] This embodiment is a further specification of the speech signal reconstruction process in claim 1; in response to signal similarity Greater than the preset coupling threshold The system is based on known theoretical distortion characteristics. Calculate the inverse filter parameters In this process, this embodiment uses Wiener filtering to construct an inverse filter in order to address the power spectrum of the reference speech. To address the issue of filter coefficient divergence caused by near-zero noise levels in the silent segment, a lower bound constraint mechanism was introduced. Simultaneously, to ensure the filter can suppress both motor self-noise and ambient background noise, the noise term in the denominator employs a full noise power model, calculated as follows: ; in, The source is a calculated value; its physical meaning is an inverse filter parameter, which is dimensionless. The source is the time-varying transfer function calculated based on the multipath differential Doppler model in Example 3, which varies with time frames. Dynamically updated; superscript : indicates complex conjugate operation, used to correct phase frequency characteristics; The source is synthetic calculation, and its physical meaning is the total noise power spectrum of the current frame, in units of... The system will use the current time corresponding to Example 3. self-noise power Compared with the background noise model power in Example 4 Performing incoherent superposition, i.e. ; The source is based on standard semantic reference features. Calculated, i.e. Get the current frame The corresponding feature slices, in physical terms, represent the hypothetical pure speech power spectrum; The source is a preset minimum power reference value, for example... This ensures that the reciprocal of the signal-to-noise ratio in the denominator is always bounded, thereby guaranteeing the numerical convergence of the inverse filter parameters across the entire frequency band. The system will calculate Applied to raw acoustic signals Considering Due to its rapidly changing characteristics, to avoid inter-frame discontinuities (i.e., click noise) during temporal synthesis, this embodiment explicitly employs a weighted overlapping summation (WOLA) architecture: Analysis window processing: for the original acoustic signal Perform frame segmentation and multiply each frame by an analysis window function. In this embodiment, the Hanning window is used to suppress spectral leakage; frequency domain filtering: an FFT is performed on the windowed frame, and the result is compared with the current frame. Perform pointwise complex multiplication; Composite window processing and superposition: Perform IFFT transform on the filtered result back to the time domain, and multiply again by the composite window function. In this embodiment, the Hanning window is also used, and the frames are accumulated to the corresponding positions in the output buffer according to a preset frame shift, such as 50% overlap. Through the above processing flow that satisfies the constant overlap addition COLA constraint, the system can remove non-stationary noise and reverberation caused by robot motion to the greatest extent, thereby restoring a clean speech signal.

[0022] Example 4: The multidimensional data acquisition module includes: a microphone array and a motion parameter detection unit; The microphone array is used to acquire raw acoustic signal data, which includes time-domain waveform data and frequency-domain feature data. The motion parameter detection unit is used to collect kinematic telemetry data, which includes chassis speed, acceleration, head servo angle, and motor PWM duty cycle.

[0023] This embodiment is a further specification of the hardware configuration of the multi-dimensional data acquisition module in claim 3. The system is configured with a microphone array, typically a ring array consisting of 4-6 microphones, with a sampling rate set to 16kHz or 48kHz, to acquire raw acoustic signal data containing PCM format time-domain waveform data and frequency-domain feature data after FFT transformation. Simultaneously, the system integrates a motion parameter detection unit, which consists of a chassis MCU and an IMU sensor, to acquire kinematic telemetry data. Specifically, this data includes chassis speed and acceleration for calculating Doppler frequency shift, head servo angle for real-time updating of the sound source relative to the microphone coordinate transformation matrix, and motor PWM duty cycle as the key value for querying the motor noise harmonic library. This embodiment clarifies the physical source and hardware foundation of the data, ensuring the completeness and availability of the input parameters for subsequent simulation models, and providing solid physical data support for high-precision acoustic simulation.

[0024] Example 5: The system also includes: a control execution module; The control execution module receives voice commands from the signal coupling decision module and controls the robot to perform corresponding service actions based on the voice commands.

[0025] This embodiment further supplements the execution phase of the system of claim 1; on this basis, the system also integrates a control execution module, which is usually deployed in the robot's main control computer, such as a ROS-based navigation board; the control execution module receives the restored voice commands output by the signal coupling decision module, such as text commands in the form of stop moving or go to the kitchen; This module converts the aforementioned voice commands into low-level control commands for the robot, such as Twist speed messages or task scheduling commands. Specifically, the control execution module maintains a semantic-action mapping table, which stores the correspondence between the voice command text key and the ROSActionGoalValue. For example, when a stop movement command is received, the system looks up the table to match the / move_base / cancel topic and issues an interrupt signal; when a go to the kitchen command is received, the system extracts the preset kitchen coordinates. The instructions are then encapsulated into MoveBaseGoal messages and sent to the navigation stack, thereby enabling deterministic action execution; the robot then executes the corresponding physical service actions based on these instructions. This embodiment realizes a complete closed loop from acoustic perception to physical action, ensuring that the recognized voice commands can be transformed into actual service value in real time, and improving the robot's responsiveness in complex interactive scenarios.

[0026] Example 6: The dynamic simulation processing module is also used for: When the robot's movement velocity vector changes, the dynamic transfer function is updated in real time to adjust the generation parameters of the theoretical distortion features.

[0027] This embodiment further specifies the real-time mechanism of the dynamic simulation processing module in claim 3; the system is configured with an event triggering mechanism for monitoring kinematic telemetry data; the system detects the moving velocity vector in real time. The rate of change; in response to the rate of change exceeding a preset threshold, for example The system immediately triggers the dynamic transfer function. Update calculation; updated Immediately applied to theoretical distortion characteristics The generation process adjusts the generation parameters; this means that if the robot suddenly accelerates or turns while listening to instructions, the theoretical distortion template inside the system will also deform in milliseconds to match the signal characteristics that change in reality. This embodiment ensures strict synchronization between the simulation model and physical reality. Especially in non-stationary dynamic scenarios such as sudden stops or turns of the robot, this real-time update mechanism is the key to ensuring that the system does not lose target instructions, and effectively solves the problem of recognition failure caused by the inability of traditional algorithms to adapt to rapid changes in the sound field.

[0028] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent service robot system integrating voice recognition, characterized in that, include: Multidimensional data acquisition module, dynamic simulation processing module, and signal coupling decision module; The multidimensional data acquisition module is used to acquire the original acoustic signal data of the robot's current environment and the kinematic telemetry data of the robot, and send the original acoustic signal data and the kinematic telemetry data to the dynamic simulation processing module. The dynamic simulation processing module is used to construct standard semantic reference features based on the current interaction context, and combine the kinematic telemetry data to perform acoustic channel distortion simulation on the standard semantic reference features in order to generate theoretical distortion features. The signal coupling decision module is used to extract real signal features from the original acoustic signal data and calculate the signal similarity between the real signal features and the theoretical distortion features; The configuration is as follows: if the signal similarity is greater than a preset coupling threshold, the original acoustic signal data is determined to contain a valid voice command, and the original acoustic signal data is restored based on the theoretical distortion characteristics to output the voice command; otherwise, the original acoustic signal data is determined to be environmental noise and a filtering operation is performed.

2. The intelligent service robot system with integrated speech recognition according to claim 1, characterized in that, The dynamic simulation processing module is also used for: Based on the current interaction context, candidate keywords are extracted; Generate standard acoustic feature vectors for the candidate keywords in a preset anechoic chamber environment, which serve as the standard semantic reference features.

3. The intelligent service robot system with integrated speech recognition according to claim 2, characterized in that, The dynamic simulation processing module is also used for: Based on the kinematic telemetry data, the motor operating state parameters and movement speed vector of the robot are analyzed. Based on the motor operating state parameters, a preset motor noise harmonic library is queried to synthesize the self-noise spectrum; Based on the moving velocity vector and the preset environmental prior map, the dynamic transfer function from the sound source to the microphone is calculated; The standard semantic reference features are convolved or processed using the dynamic transfer function, and the self-noise spectrum is superimposed to generate the theoretical distortion features.

4. The intelligent service robot system with integrated speech recognition according to claim 1, characterized in that, The signal coupling decision module is also used for: Obtain a preset background noise reference model; The difference features between the original acoustic signal data and the background noise reference model are calculated as the real signal features; Calculate the difference features between the theoretical distortion features and the standard semantic reference features, and use them as theoretical residual features; The cross-correlation coefficient or structural similarity index between the actual signal features and the theoretical residual features on the time-frequency spectrum is calculated as the signal similarity.

5. The intelligent service robot system with integrated speech recognition according to claim 1, characterized in that, The signal coupling decision module is also used for: If it is determined that the original acoustic signal data contains valid speech commands, the corresponding inverse filter parameters are calculated based on the theoretical distortion characteristics. The inverse filtering parameters are applied to the original acoustic signal data to restore a clean speech signal.

6. The intelligent service robot system with integrated speech recognition according to claim 3, characterized in that, The multidimensional data acquisition module includes: a microphone array and a motion parameter detection unit; The microphone array is used to acquire the raw acoustic signal data, which includes time-domain waveform data and frequency-domain feature data. The motion parameter detection unit is used to collect the kinematic telemetry data, which includes chassis speed, acceleration, head servo angle, and motor PWM duty cycle.

7. The intelligent service robot system with integrated speech recognition according to any one of claims 1-6, characterized in that, The system also includes: a control execution module; The control execution module is used to receive the voice command output by the signal coupling decision module, and control the robot to perform corresponding service actions according to the voice command.

8. The intelligent service robot system with integrated speech recognition according to claim 3, characterized in that, The dynamic simulation processing module is also used for: When the robot's movement velocity vector changes, the dynamic transfer function is updated in real time to adjust the generation parameters of the theoretical distortion feature.

Images