Voice control method for intelligent detection instrument in coating laboratory

Abstract

The invention relates to a voice control method for an intelligent detection instrument in a coating laboratory. The voice control method comprises the following steps: firstly, acquiring and preprocessing voice; acquiring a voice signal through audio input equipment; converting voice signals into character information by using a voice recognition model, detecting wake-up words, detecting whether the voice contains preset wake-up words or not, triggering system response, performing intention recognition, performing intention recognition on the converted characters, analyzing operation intention and parameters of a user, and outputting the converted characters; the intention recognition result is sent to the upper computer for operation execution, and then voice feedback is conducted. According to the technical scheme, the natural language interaction control of the detection instrument is realized through the complete process of voice-to-text conversion, intention recognition and TTS voice feedback; a hot word management and wake-up mechanism is adopted, so that the recognition accuracy in a professional environment is improved; and an interruption mechanism and real-time response are supported, and the user experience is improved.

Description

Technical Field

[0001] This invention relates to the field of voice recognition and instrument control technology, specifically to a method for voice control of an intelligent testing instrument in a paint laboratory. Background Technology

[0002] In a coatings laboratory testing environment, operators typically need to manually operate various testing instruments, such as scrub resistance testers, viscometers, colorimeters, and gloss meters. Traditional operating methods have the following problems: Operators need to frequently come into contact with the instrument interface, which may lead to sample contamination or operation interruption. During the experiment, the operator's hands may be occupied, making it impossible to operate the instrument in a timely manner; Existing voice control solutions have shortcomings in recognizing technical terms.

[0003] Existing voice control technologies are mostly focused on general scenarios, and there is a lack of targeted solutions for professional environments such as paint laboratories.

[0004] Existing solutions, such as those using APIs developed with iFlytek's speech recognition technology, require online access. In real-world laboratory environments, poor network signals can lead to suboptimal speech recognition performance. Furthermore, online speech recognition solutions incur fees based on API calls, and the presence of other sounds in the laboratory can trigger additional API calls and incur additional charges, significantly increasing the cost of equipment use. Therefore, developing a voice-controlled method for intelligent coating laboratory testing instruments is essential. This method should be offline and require no additional payment from the user. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a voice control method for a coating laboratory testing instrument, which realizes non-contact intelligent control of the testing instrument through voice recognition, intent recognition and voice broadcast.

[0006] This invention provides the following technical solution: a method for voice control of an intelligent testing instrument in a paint laboratory, comprising: firstly, voice acquisition and preprocessing: acquiring voice signals through an audio input device; then, speech-to-text conversion: using a speech recognition model to convert the voice signal into text information. The voice acquisition module receives the user's original voice commands, acquires mono voice signals at a fixed sampling rate through an audio input device such as a microphone, and calls the sounddevice library to read the audio stream, and uses the silero_vad library for voice activity detection (VAD), setting an appropriate detection threshold to accurately determine the start and end of the speech. The speech recognition module converts the voice signal into text, detects wake words, and inputs the acquired audio data into a Paraformer speech recognition model based on the FunASR framework to convert the speech into text. Simultaneously, the system checks whether the recognition result contains a preset wake word (such as "Xiaowei Xiaowei"). If a wake word is detected, the system is activated and the last wake time (lastWakeTime) is updated; if no wake word is detected, listening continues.

[0007] As a preferred embodiment of the present invention, the wake-up word is then detected: whether the speech contains a preset wake-up word, and the system response is triggered.

[0008] As a preferred embodiment of the present invention, intent recognition is then performed: the converted text is subjected to intent recognition, parsing the user's operational intent and parameters. The intent recognition module understands the intent of the user's text command and extracts key parameters. The text output by the speech recognition module is sent to the intent recognition service (e.g., http: / / 127.0.0.1:5000 / intent) via an HTTP POST request. This service uses a JointBERT-based model for joint intent classification and slot filling, outputting structured JSON data containing intents (e.g., "parameter settings") and key parameters (e.g., number of flushes, flushing speed, etc.).

[0009] As a preferred embodiment of the present invention, the instruction is then executed: the intent recognition result is sent to the host computer to perform the operation. The instrument operation is controlled to perform corresponding actions according to the parsed intent. After the voice recognition module receives the JSON data returned by the intent recognition server, it forwards it to the host computer software through a virtual serial port (such as COM66, baud rate 9600). After receiving the JSON data, the host computer parses it and executes the specific operation, such as starting the washability test.

[0010] As a preferred embodiment of the present invention, voice feedback is then provided: the operation result or status information is fed back to the user in voice form through TTS technology, and the TTS voice output feeds back the system status or operation result to the user in voice form. When a wake word is detected, the system immediately responds with "I am here" via TTS (using Microsoft SpeechSynthesizer for TTS voice synthesis). Operation confirmation feedback: before and after the "perform instrument operation" step, "starting test" can be fed back as needed. Sleep notification: if there are no new instructions within 15 seconds after the system is activated, it will automatically enter sleep mode and notify the user via TTS.

[0011] As a preferred embodiment of the present invention, the speech-to-text step employs the Paraformer speech recognition model and combines it with Silero VAD for speech activity detection to improve recognition accuracy.

[0012] As a preferred embodiment of the present invention, the intent recognition step adopts a joint intent recognition and slot filling model based on JointBERT, which can simultaneously identify user intent and extract key parameters.

[0013] As a preferred embodiment of the present invention, the system supports hot word management, which can dynamically add and delete hot words based on the professional terminology of the coating laboratory, thereby improving the professional terminology recognition rate.

[0014] The beneficial effects of this invention are: this technical solution realizes natural language interactive control of the detection instrument through a complete process of speech-to-text, intent recognition and TTS voice feedback; it improves the recognition accuracy in professional environments by adopting hot word management and wake-up mechanism; and it supports interruption mechanism and real-time response, thereby enhancing the user experience.

[0015] A voice-controlled method for intelligent testing instruments in coating laboratories is essential, and this method can be used offline without requiring a network; moreover, it does not require users to pay extra. In contrast, current offline solutions on the market require fixed input commands. For laboratory instruments with speed ranges of 0-10000 mm / s, achieving automatic speed setting for voice control requires inputting 10,000 commands in advance, which is not achievable with current general offline modules. Attached Figure Description

[0016] Figure 1 This is an overall flowchart of the present invention; Figure 2 This is a detailed flowchart of the speech recognition process of the present invention; Figure 3 This is the system structure of the present invention. Detailed Implementation

[0017] To make the technical problems solved by the present invention, the technical solutions adopted, and the technical effects achieved clearer, the technical solutions of the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1 like Figures 1 to 3 As shown, a method for voice control of an intelligent testing instrument in a paint laboratory includes: first, voice acquisition and preprocessing: acquiring voice signals through an audio input device; then, speech-to-text conversion: using a speech recognition model to convert the voice signal into text information. The voice acquisition module receives the user's original voice commands, acquires mono voice signals at a sampling rate of 16kHz through an audio input device such as a microphone, and calls the sounddevice library to read the audio stream, and uses the silero_vad library for voice activity detection (VAD), setting a detection threshold of 0.3 to accurately determine the start and end of the speech. The speech recognition module converts the voice signal into text and detects wake words, inputting the acquired audio data into a Paraformer speech recognition model based on the FunASR framework to convert the speech into text. Simultaneously, the system checks whether the recognition result contains a preset wake word (such as "Xiaowei Xiaowei"). If a wake word is detected, the system is activated and the last wake time (lastWakeTime) is updated; if no wake word is detected, listening continues. Next, wake-up word detection: Detects whether the speech contains a preset wake-up word, triggering a system response, and then performs intent recognition: Performs intent recognition on the converted text, parsing the user's operation intent and parameters. The intent recognition module understands the user's text command intent and extracts key parameters. The text output by the speech recognition module is sent to the intent recognition service (e.g., http: / / 127.0.0.1:5000 / intent) via an HTTP POST request. This service uses a JointBERT-based model for joint intent classification and slot filling, outputting structured JSON data containing intents (e.g., "parameter settings") and key parameters (e.g., number of flushes, flushing speed, etc.). The process then proceeds to command execution: the intent recognition result is sent to the host computer for operation, followed by voice feedback: the operation result or status information is fed back to the user in voice form via TTS technology. The instrument operation is controlled according to the parsed intent to perform corresponding actions. After receiving the JSON data returned by the intent recognition server, the voice recognition module forwards it to the host computer software via a virtual serial port (e.g., COM66, baud rate 9600). The host computer receives the JSON data, parses it, and executes the specific operation, such as starting a washability test. The TTS voice output feeds back the system status or operation result to the user in voice form. Operation confirmation feedback: before and after the "execute instrument operation" step, "starting test" can be fed back as needed. Sleep notification: if there are no new instructions within 15 seconds after activation, the system will automatically enter sleep mode and notify the user via TTS "I'm going to rest now". In this example, the speech-to-text step uses the Paraformer speech recognition model and combines it with SileroVAD for speech activity detection to improve recognition accuracy. In this example, the intent recognition step adopts a joint intent recognition and slot filling model based on JointBERT, which can simultaneously recognize user intent and extract key parameters. The system described in this example supports hot word management, which can dynamically add and delete hot words based on the professional terminology of the coating laboratory, thereby improving the recognition rate of professional terminology; This technical solution achieves natural language interactive control of the detection instrument through a complete process of speech-to-text, intent recognition, and TTS voice feedback; it improves the recognition accuracy in professional environments by adopting hot word management and wake-up mechanisms; and it supports interruption mechanisms and real-time responses, enhancing the user experience. In this example, a voice-controlled method for intelligent testing instruments in a paint laboratory is essential, and this method can be used offline without a network. Furthermore, it requires no additional payment from the user. In contrast, current offline solutions on the market require fixed input commands. For laboratory instruments with speed ranges of 0-10000 mm / s, achieving automatic speed setting via voice control would require pre-entering 10,000 commands, which is impossible with current general-purpose offline modules. The specific implementation of the voice control method for the coating laboratory testing instrument provided by this invention is as follows: Speech Acquisition and Preprocessing: A 16kHz sampling rate, mono speech signal was acquired using an audio input device. Silero VAD was used for speech activity detection, with a detection threshold set to 0.3. The start and end boundaries of speech were detected through real-time audio stream processing.

[0019] Speech-to-text processing: Load the speech recognition model, combine it with the hot word manager, dynamically load the professional terminology of the coating laboratory, recognize the detected speech segments, and output the text results.

[0020] Wake-up Mechanism: The system maintains a wake-up word list, supporting custom wake-up words such as "Xiao Wei Xiao Wei". The wake-up word list will be automatically added to the hot word list to ensure normal wake-up. After detecting the wake-up word, subsequent user voice input is processed. If there is no voice input within 15 seconds, it will automatically enter sleep mode to save system resources.

[0021] Intent Recognition: The JointBERT-based joint intent recognition and slot filling model is used to identify user intents (such as "start test", "stop measurement", etc.) and extract key parameters (such as parameter name and parameter value). For example, if the user wants to "set the temperature to thirty", the model will automatically extract the parameter name as "temperature" and the parameter value as "thirty".

[0022] Command generation and execution: The identified intent and parameters are sent to the host computer via serial port. The host computer parses the JSON to obtain the user's intent, executes the corresponding operation, and provides feedback to the user via voice announcement. If the intent is to set parameters, the program will automatically convert "thirty" to "30" for the actual parameter setting operation.

[0023] Voice feedback: TTS speech synthesis is performed using Microsoft SpeechSynthesizer, supporting an interruption mechanism. New speech content can interrupt the currently playing content. During playback, if the user triggers the wake word again, the system will prioritize responding to the user and announce "I am here".

[0024] Hotword Management: The system maintains a hotword file, supporting the dynamic addition and deletion of technical terms. Hotwords include instrument names, operating commands, etc.

[0025] In a specific paint laboratory scenario, the operator can say, "Start test." After speech recognition and intent parsing, the system will generate the corresponding intent recognition result and send it to the host computer software to initiate the test process. The system will then provide voice feedback, "Test started." Upon completion of the test, the system will announce the test results.

[0026] The technical solution of this invention effectively solves the problems of convenience and accuracy in instrument operation in coating laboratories, and improves experimental efficiency and data reliability.

[0027] The voice-controlled intelligent function in this invention includes: Automatic status announcement: Automatically zeroes the instrument upon startup, detects the instrument's horizontal status and rotor installation status, and provides feedback to the user via voice. Voice command control: Users can use voice commands to switch functions, start tests, generate reports, and perform other operations. It supports multiple languages, including Chinese and English, which significantly improves work efficiency. In this invention, voice control technology reduces operational complexity by minimizing human intervention, while also avoiding errors caused by manual input. In this example, voice data acquisition and recognition involves: acquiring the voice commands of the experimenters through a microphone, inputting the audio signal into a preset voice recognition model, and converting it into text data; Command Execution and Feedback: Recognized text commands trigger device operations, such as starting a test or switching rotors, and the execution results are fed back via voice or screen. The Shenghua BEVS AI viscometer supports multiple languages, including Chinese and English, and can automatically complete speed adaptation and data acquisition.

[0028] Efficiency Improvement: Voice control reduces manual operation costs by 60%, such as the Shenghua viscometer which can quickly complete the testing process through commands.

[0029] Enhanced accuracy: Avoiding errors from manual input, such as the automatic rotor recognition function, reduces human intervention.

[0030] Intelligent expansion: Some systems support fault prediction and maintenance reminders, further optimizing laboratory management.

[0031] In this invention: 1. Voice command control for instrument operation Users can directly control the testing instrument's function switching, test initiation, and report generation via voice commands. For example, Shenghua Industrial's BEVS AI viscometer supports multiple languages, including Chinese and English, allowing users to complete the entire process from testing to reporting through voice commands, reducing manual intervention.

[0032] 2. Real-time voice recording of experimental data. The technology converts voice input during experiments into text data and automatically associates it with experimental values ​​to generate reports. In Shanghai Jianke Testing's patented technology, voice data is processed by a language recognition model, text entity features are extracted, and specific experimental values ​​are matched to ultimately generate standardized experimental record reports.

[0033] 3. Voice Interaction and Intelligent Navigation By combining Large Language Model (LLM) technology, natural language-driven business system interactions can be achieved. For example, 3D World's intelligent voice interaction solution supports voice commands to trigger instrument operation, data entry, and report modification, and improves operational convenience through a multimodal semantic understanding engine.

[0034] 4. Voice-assisted calibration and status prompts Some instruments automatically provide voice prompts regarding calibration status and rotor installation during power-on or testing. For example, the BEVSAI viscometer automatically zeros upon power-on and provides voice feedback on the instrument's status, ensuring operational accuracy.

[0035] Technological advantages Efficiency Improvement: Voice control reduces manual operation time; for example, the viscosity testing process can be shortened by more than 30%.

[0036] Enhanced accuracy: Avoids errors in manual recording, improving the integrity of experimental data.

[0037] Multi-scenario adaptation: Supports speech recognition in complex laboratory environments, such as instruction filtering in noisy backgrounds.

[0038] This invention employs a multi-microphone array: a voice array with a 6-microphone ring layout, combined with beamforming technology to achieve 120° sound source localization, and applies deep learning noise reduction algorithms (such as CRN networks), which can maintain a wake-up rate of over 95% in an 80dB noise environment.

[0039] Isolate experimental noise: For common equipment operating noises in coating laboratories (such as mixers and centrifuges), it is necessary to reduce interference through physical sound insulation or active noise reduction technology.

[0040] Multimodal fusion: Combining multimodal interactions such as voice and gestures improves recognition robustness in complex scenarios. For example, gestures can be used to assist in confirming the intent of a command, reducing misrecognition.

[0041] Domain-specific models: For coating-related technical terms (such as "viscosity meter" and "adhesion test"), a dedicated language model needs to be trained to avoid performance degradation of the general model in professional scenarios.

[0042] Integration with LIMS systems: Integrating the voice recognition system with the Laboratory Information Management System (LIMS) enables automatic association of test data and report generation. For example, Shanghai Jianke's patented technology generates test reports directly through voice input, improving data integrity.

[0043] Real-time feedback mechanism: The recognition results need to be fed back in real time via voice or screen, supporting multiple rounds of interactive correction. For example, the BEVS AI viscometer confirms the detection values ​​through voice interaction, reducing manual input errors.

[0044] Iterative training with data: Regularly collect laboratory speech data to optimize the model's adaptability to specific accents and speech rates.

[0045] Equipment maintenance early warning: Utilize AI to predict the failure risk of equipment such as microphones and perform timely maintenance to ensure recognition stability.

[0046] The above measures can significantly improve the accuracy and efficiency of voice recognition in coating laboratories and promote the intelligentization of the testing process.

[0047] In this example, pre-trained models such as Whisper are used, trained with 680,000 hours of labeled data. This supports multilingual and professional terminology recognition. Fine-tuning is performed on specific terms in the coatings industry to improve the model's accuracy in recognizing professional vocabulary. The MoME (Mixture of Matryoshka Experts) framework is adopted to combine audio and visual information (such as lip shape) to improve recognition accuracy. In noisy environments, this dual perception capability can significantly improve the recognition effect. Tools such as PaddleSpeech are used to achieve mixed Chinese and English recognition, and cross-language processing is optimized through the Conformer architecture and WFST framework.

[0048] Example 2: Corresponding Appendix Figure 1 Overall process description Figure 1 The overall control flow of the method of the present invention is shown, and the specific steps are as follows: Voice acquisition module Function: Receives raw voice commands from the user.

[0049] Implementation: A mono audio signal is acquired at a sampling rate of 16kHz using an audio input device such as a microphone. This module calls the sounddevice library to read the audio stream and uses the silero_vad library for speech activity detection (VAD), setting the detection threshold to 0.3 to accurately determine the start and end of speech.

[0050] Speech recognition module Function: Converts speech signals into text and detects wake words.

[0051] Implementation: The acquired audio data is input into the Paraformer speech recognition model based on the FunASR framework to convert speech into text. Simultaneously, the system checks whether the recognition result contains a preset wake word (e.g., "Xiaowei Xiaowei"). If a wake word is detected, the system is activated and the last wake-up time (lastWakeTime) is updated; if no wake word is detected, listening continues.

[0052] Intent recognition module Function: Understand the intent of user text commands and extract key parameters.

[0053] Implementation: The text output by the speech recognition module is sent to the intent recognition service (e.g., http: / / 127.0.0.1:5000 / intent) via an HTTP POST request. This service uses a JointBERT-based model for joint intent classification and slot filling, outputting structured JSON data containing the intent (e.g., "parameter settings") and key parameters (e.g., number of flushes, flushing speed, etc.).

[0054] Perform instrument operation Function: Control the instrument to perform corresponding actions based on the interpreted intent.

[0055] Implementation: After receiving the JSON data returned by the intent recognition server, the voice recognition module forwards it to the host computer software through a virtual serial port (such as COM66, baud rate 9600). After receiving the JSON data, the host computer parses it and executes specific operations, such as starting a washability test.

[0056] TTS voice output Function: Provides system status or operation results to the user in voice format.

[0057] Implementation method: This step is implemented through the above-mentioned multiple stages: Wake-up feedback: When a wake-up word is detected, the system immediately responds with "I am here" via TTS (using Microsoft SpeechSynthesizer for TTS speech synthesis).

[0058] Operation confirmation feedback: Before and after the "Perform instrument operation" step, feedback such as "Testing is starting" can be provided as needed.

[0059] Hibernation notification: If there are no new commands within 15 seconds of activation, the system will automatically enter hibernation and notify you via TTS that "I'm going to rest now".

[0060] Example 3: Corresponding Appendix Figure 2 Detailed explanation of the speech recognition process Figure 2 Detailed Explanation Figure 1 The internal processing logic from "voice acquisition" to "voice recognition" in the middle is as follows: System initialization: Load the speech recognition model (Paraformer), the speech activity detection model (Silero VAD), and initialize the audio stream parameters.

[0061] Start audio stream: Start the audio input stream, set the sampling rate to 16kHz, and begin looping through audio data blocks (CHUNK_SIZE = 512).

[0062] VAD speech detection: Analyzes each read audio data block in real time to determine if it contains valid speech. When the speech energy exceeds a preset threshold (0.3), it is determined that speech has started.

[0063] Cache audio data: From the moment the voice is detected, not only is the current audio data cached, but also the previously saved 200 milliseconds of historical audio data (PREV_CHUNKS) is added to ensure the integrity of the voice segment.

[0064] Continuous Acquisition and Termination Detection: Audio is continuously acquired until the VAD detects the end of the speech. At this point, all cached audio data (historical data + current speech data) is concatenated into a complete speech segment.

[0065] Perform speech recognition: Feed the complete speech segment into the Paraformer model for recognition, and finally output the text result.

[0066] Wake word detection and state management: Check if the identified text is a preset wake word.

[0067] If it is a wake word, the system is activated, the wake-up time is updated, and the TTS voice output "I am here" is triggered directly for feedback. Then the process returns to continue listening, without sending the wake word to the intent recognition.

[0068] If it is not a wake word, check if the system is within the wake-up validity period (e.g., within 15 seconds of the last wake-up). If it is within the validity period, send the text to the intent recognition; if it has expired, the process returns to continue listening.

[0069] Subsequent process: After intent recognition, the instrument operation is executed, and finally the TTS voice output of the operation result is triggered.

[0070] Example 4: Corresponding Appendix Figure 3 System architecture description Figure 3 The layered architecture of the hardware and software system implementing this method is shown, with the functions of each layer as follows: User interaction layer Components: microphone, speaker.

[0071] Function: As the physical interface between the system and the user, it is responsible for receiving voice commands and playing voice feedback.

[0072] Speech processing layer Components: Audio preprocessing module, speech recognition module (Paraformer).

[0073] Function: Processes raw audio signals to convert speech to text. VAD detection is used for endpoint detection, and the Paraformer model is responsible for high-precision recognition.

[0074] Intelligent Analysis Layer Components: Intent recognition module (JointBERT), slot extraction module (JointBERT).

[0075] Functions: Perform deep semantic analysis on the identified text; the intent recognition module determines the user's intent, while the slot extraction module extracts key parameters (such as the operation object, value, etc.) from the intent to form structured operation instructions.

[0076] Control Execution Layer Component: Instrument operation module.

[0077] Function: Receives structured instructions from the intelligent analysis layer and drives the host computer of the coating testing instrument to complete the specified actions through a virtual serial port.

[0078] Feedback layer Component: TTS voice output module.

[0079] Function: This layer is crucial for the closed-loop system interaction. It receives execution feedback from the control execution layer or directly responds to wake-up events from the voice processing layer, calls the TTS engine (Microsoft SpeechSynthesizer) to generate a voice audio stream, and pushes it to the speaker in the user interaction layer for playback. Its diverse input sources ensure a smooth and user-friendly interaction process.

[0080] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0081] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for voice control of an intelligent testing instrument in a paint laboratory, characterized in that, It includes first voice collection and preprocessing: collecting voice signals through an audio input device; then performing voice-to-text conversion: using a speech recognition model to convert voice signals into text information. The voice collection module is used to receive the user's original voice instructions, collect monophonic voice signals at a fixed sampling rate through an audio input device such as a microphone, and this module calls the sounddevice library to read the audio stream and uses the silero_vad library for voice activity detection (VAD), setting a suitable detection threshold to accurately judge the start and end of the voice. The speech recognition module converts the voice signal into text, detects the wake word, and inputs the collected audio data into the Paraformer speech recognition model based on the FunASR framework to convert the voice into text. At the same time, the system checks whether the recognition result contains a preset wake word (such as "Xiaowei Xiaowei"). If the wake word is detected, the system is activated and the last wake time (lastWakeTime) is updated; if the wake word is not detected, the system continues to listen.

2. The method for voice control of an intelligent testing instrument in a coating laboratory according to claim 1, characterized in that, Subsequently, wake word detection: detecting whether the voice contains a preset wake word to trigger the system response.

3. The method for voice control of an intelligent testing instrument in a coating laboratory according to claim 1, characterized in that, Then, intent recognition: performing intent recognition on the converted text to parse the user's operation intent and parameters. The intent recognition module understands the intent of the user's text instructions and extracts key parameters, and sends the text output by the speech recognition module to the intent recognition server (such as: http: / / 127.0.0.1:5000 / intent) in the form of an HTTP POST request. This server uses a model based on JointBERT for joint intent classification and slot filling, and outputs structured JSON data, including the intent (such as "parameter setting") and key parameters (such as the number of brushing times, brushing speed, etc.).

4. The method for voice control of an intelligent testing instrument in a coating laboratory according to claim 1, characterized in that, Then, instruction execution: sending the intent recognition result to the host computer to perform operations. The execution instrument operation controls the instrument to perform corresponding actions according to the parsed intent. After the speech recognition module receives the JSON data returned by the intent recognition server, it forwards it to the host computer software through a virtual serial port (such as COM66, baud rate 9600). After receiving the JSON data, the host computer analyzes and executes specific operations, such as starting the washability test.

5. The method for voice control of an intelligent testing instrument in a coating laboratory according to claim 1, characterized in that, After executing the corresponding instruction, voice feedback is provided: the operation result or status information is fed back to the user in the form of voice through TTS technology. When the wake word is detected, the system immediately provides a TTS feedback of "I'm here" (using Microsoft SpeechSynthesizer for TTS voice synthesis). Operation confirmation feedback: Before and after the "execution instrument operation" step, "The test is starting" can be fed back as needed. Sleep notification: When there is no new instruction within 15 seconds after the system is activated, it will automatically enter the sleep state and notify the user through TTS.

6. The method for voice control of an intelligent testing instrument in a coating laboratory according to claim 1, characterized in that, The voice-to-text step uses the Paraformer speech recognition model and combines Silero VAD for voice activity detection to improve the recognition accuracy.

7. The method for voice control of an intelligent testing instrument in a coating laboratory according to claim 1, characterized in that, The intent recognition step employs a joint intent recognition and slot filling model based on JointBERT, which can simultaneously identify user intent and extract key parameters.

8. The method for voice control of an intelligent testing instrument in a coating laboratory according to claim 1, characterized in that, The system supports hot word management, which can dynamically add and delete hot words based on the professional terminology of the coating laboratory, thereby improving the recognition rate of professional terminology.

Images