Method and system for automatically adapting to television far-field voice wake-up model

Abstract

The invention discloses a method and system for automatically adapting to a television far-field voice wake-up model, and the method comprises the steps: obtaining an audio, determining the wake-up rate of each far-field voice wake-up model for the audio through a wake-up rate simulation tool, and obtaining wake-up rate data; based on the wake-up rate data, obtaining a target far-field voice wake-up model; and receiving a target APK file of the target far-field voice wake-up model, and completing installation of the target APK file. According to the method, the target far-field voice wake-up model is evaluated and screened out by using the wake-up rate simulation tool based on the acquired audio, and the installation of the APK file corresponding to the model is completed, so that the problems of high time cost and low efficiency caused by repeated physical equipment adjustment and dependence on third-party debugging in a traditional voice model adaptation method are solved; and the optimization efficiency and quality of the voice interaction function are obviously improved.

Description

Technical Field

[0001] The present invention relates to the field of voice interaction technology, and in particular to a method and system for automatically adapting a television far-field voice wake-up model. Background Art

[0002] With the rapid popularization of voice interaction technology, far-field voice interaction has become an important part of consumer electronic products such as TVs and smart speakers, significantly improving the user's convenience and experience. Most mid-to-high-end TVs on the market have integrated this function, allowing users to interact with devices through natural language. However, although far-field voice interaction has brought improvements in user experience, its adaptation process is very complicated and costly. This is because the adaptation of the voice interaction solution not only involves structural design and acoustic optimization, but also needs to overcome the structural resonance noise generated when the TV and the speaker sound, as well as the influence of the speaker position on the voice pickup effect. In order to ensure the best voice recognition performance, the above physical and acoustic factors must be adjusted to the ideal state and professionally debugged by a third-party voice algorithm supplier. This usually means that binary model resources in a specific format (such as .bin) need to be output, then compiled into APK and installed in the TV system to realize the voice interaction function. Since the adaptation process involves the intervention of a third party, including the delivery of the machine back and forth to complete the debugging work, the time cost of the whole process is extremely high. In addition, troubleshooting for each individual model (such as modifying the mold or adjusting the speaker position) also increases development time and cost, resulting in inefficiency and poor model compatibility. These limitations hinder the rapid promotion and application of far-field voice interaction technology on more TV models.

[0003] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for automatically adapting a TV far-field voice wake-up model in response to the above-mentioned defects of the prior art, aiming to solve the problems of high cost, long cycle and low efficiency when traditional TVs adapt to far-field voice interaction solutions.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides a method for automatically adapting a television far-field voice wake-up model, wherein the method comprises:

[0007] Acquire audio, and use a wake-up rate simulation tool to determine the wake-up rate of each far-field voice wake-up model for the audio, to obtain wake-up rate data;

[0008] Based on the wake-up rate data, a target far-field voice wake-up model is obtained;

[0009] Receive the target APK file of the target far-field voice wake-up model, and complete the installation of the target APK file.

[0010] In one implementation, obtaining audio includes:

[0011] Acquire a test tool, and determine a test scenario based on the test tool;

[0012] Based on the test scenario, the audio corresponding to the test scenario is obtained.

[0013] In one implementation, the using of a wake-up rate simulation tool to determine the wake-up rate of each far-field voice wake-up model to the audio to obtain the wake-up rate data includes:

[0014] Using a wake-up rate simulation tool, preprocessing the audio to obtain processed audio, wherein the preprocessing includes any one or more of echo cancellation, noise reduction, dereverberation, and gain adjustment;

[0015] Based on the processed audio, a wake-up rate simulation tool is used to determine the wake-up rate of each far-field voice wake-up model to the processed audio to obtain wake-up rate data.

[0016] In one implementation, based on the processed audio, determining the wake-up rate of each far-field voice wake-up model to the processed audio using a wake-up rate simulation tool to obtain the wake-up rate data includes:

[0017] Based on the processed audio, using a wake-up rate simulation tool to determine a score of each far-field speech wake-up model on the processed audio;

[0018] Based on the score, the wake-up rate of each far-field voice wake-up model for the processed audio is determined to obtain wake-up rate data.

[0019] In one implementation, determining the wake-up rate of each far-field voice wake-up model for the processed audio based on the score to obtain the wake-up rate data includes:

[0020] Based on the score and the preset threshold, determining the number of successful wake-up times for each far-field voice wake-up model whose score is equal to or greater than the preset threshold;

[0021] Based on the number of successful wake-up times and the total number of scoring times of each far-field voice wake-up model, the wake-up rate of each far-field voice wake-up model for the processed audio is determined to obtain wake-up rate data.

[0022] In one implementation, obtaining a target far-field voice wake-up model based on the wake-up rate data includes:

[0023] The wake-up rate simulation tool is used to sort the wake-up rate data through recursive sorting, and the far-field voice wake-up model with the highest wake-up rate is screened out to obtain the target far-field voice wake-up model.

[0024] In one implementation, the receiving of the target APK file of the target far-field voice wake-up model and completing the installation of the target APK file includes:

[0025] Based on a software development kit, compile the source code of the application of the target far-field voice wake-up model into a bytecode file, and package the bytecode file and the resource file into an APK file;

[0026] Based on the software development kit, the APK file is aligned and optimized, and the APK file is signed to obtain a target APK file.

[0027] In a second aspect, an embodiment of the present invention further provides a system for automatically adapting a TV far-field voice wake-up model, wherein the system comprises:

[0028] A wake-up rate data acquisition module is used to acquire audio and use a wake-up rate simulation tool to determine the wake-up rate of each far-field voice wake-up model for the audio to obtain wake-up rate data;

[0029] A target far-field voice wake-up model acquisition module, used to obtain a target far-field voice wake-up model based on the wake-up rate data;

[0030] An installation module is used to receive the target APK file of the target far-field voice wake-up model and complete the installation of the target APK file.

[0031] In the third aspect, an embodiment of the present invention further provides a terminal, wherein the terminal includes a memory, a processor, and an automatic adaptation TV far-field voice wake-up model program stored in the memory and runnable on the processor, and when the processor executes the automatic adaptation TV far-field voice wake-up model program, the steps of the automatic adaptation TV far-field voice wake-up model method of any one of the above-mentioned schemes are implemented.

[0032] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein a program for automatically adapting to the TV far-field voice wake-up model is stored on the computer-readable storage medium, and when the program for automatically adapting to the TV far-field voice wake-up model is executed by a processor, the steps of the method for automatically adapting to the TV far-field voice wake-up model described in any one of the above-mentioned schemes are implemented.

[0033] Beneficial effects: The present invention provides a method for automatically adapting a TV far-field voice wake-up model. Compared with the prior art, the present invention first obtains audio to provide a standardized and repeatable audio data set for subsequent model evaluation, ensuring that different far-field voice wake-up models are compared under the same conditions. Then, the wake-up rate simulation tool is used to determine the wake-up rate of each far-field voice wake-up model for the audio, and the wake-up rate data is obtained, allowing developers to quickly and accurately evaluate the performance of multiple models without relying on physical devices, greatly improving the screening efficiency and accuracy, while reducing the high time cost and resource waste caused by physical debugging. Subsequently, based on the wake-up rate data, the target far-field voice wake-up model is obtained, which improves the scientificity and accuracy of model selection, enhances the possibility of model reuse, and reduces the development cost and time required for adapting new models. Then, the target APK file of the target far-field voice wake-up model is received, and the installation of the target APK file is completed, achieving seamless connection from model selection to application deployment, simplifying the development process, and improving work efficiency. The method proposed in the present invention effectively solves the problems of frequent third-party intervention, high time cost for machine delivery, and low adaptation efficiency of a single model in the adaptation process of traditional voice interaction solutions by introducing simulation screening technology. The proposed method not only optimizes the far-field voice interaction performance and improves the reuse rate of the model, but also significantly shortens the adaptation cycle and reduces development costs. Overall, this method greatly improves the development efficiency and quality of voice interaction functions, and provides a more efficient and convenient voice interaction solution for devices such as smart TVs. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flowchart of a specific implementation method of the method for automatically adapting a TV far-field voice wake-up model provided in an embodiment of the present invention.

[0035] Figure 2 A flowchart of a preferred embodiment of the method for automatically adapting a television far-field voice wake-up model provided in an embodiment of the present invention.

[0036] Figure 3 It is a principle block diagram of a system for automatically adapting a TV far-field voice wake-up model provided by an embodiment of the present invention.

[0037] Figure 4 This is a block diagram of the internal structure principle of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0039] In the field of smart home, far-field voice interaction is one of the key technologies to improve user experience. However, traditional implementation methods usually rely on complex structural design and fine acoustic adjustment to ensure the accuracy of voice recognition and response quality. Although these measures help improve performance, they also bring many challenges: for example, the adaptation process may become extremely cumbersome due to factors such as structural resonance noise generated when the TV and the speaker make sound, and the position of the speaker. This not only prolongs the development cycle and increases costs, but also requires individual optimization of each model, such as adjusting the mold or changing the position of the speaker, resulting in high time cost, low work efficiency, and poor compatibility between different models. In addition, in the current far-field voice interaction TV solution, third-party voice algorithm suppliers are often required to participate in debugging for specific models. This process further exacerbates the above problems, making the overall solution highly customized and lacking in versatility.

[0040] In order to solve the above problems, this embodiment provides a pulse wave monitoring method based on telemetry photoplethysmography. In specific implementation, this embodiment first obtains audio to provide a standardized and repeatable audio data set for subsequent model evaluation, ensuring that different far-field voice wake-up models are compared under the same conditions. Then, the wake-up rate simulation tool is used to determine the wake-up rate of each far-field voice wake-up model for the audio, and the wake-up rate data is obtained, allowing developers to quickly and accurately evaluate the performance of multiple models without relying on physical devices, greatly improving the screening efficiency and accuracy, while reducing the high time cost and resource waste caused by physical debugging. Subsequently, based on the wake-up rate data, the target far-field voice wake-up model is obtained, which improves the scientificity and accuracy of model selection, enhances the possibility of model reuse, and reduces the development cost and time required for adapting to new models. Then, the target APK file of the target far-field voice wake-up model is received, and the installation of the target APK file is completed, realizing seamless connection from model selection to application deployment, simplifying the development process, and improving work efficiency. The method proposed in the present invention effectively solves the problems of frequent third-party intervention, high time cost for machine delivery, and low adaptation efficiency of a single model in the adaptation process of traditional voice interaction solutions by introducing simulation screening technology. The proposed method not only optimizes the far-field voice interaction performance and improves the reuse rate of the model, but also significantly shortens the adaptation cycle and reduces development costs. Overall, this method greatly improves the development efficiency and quality of voice interaction functions, and provides a more efficient and convenient voice interaction solution for devices such as smart TVs.

[0041] For example, suppose a TV manufacturer wants to quickly and efficiently integrate far-field voice interaction functions for its new model of smart TV. The method proposed in the present invention can be used. First, audio is obtained, and the audio represents audio samples of different environmental conditions and user pronunciation habits. These audio samples are used to simulate various situations that may be encountered in the real world, so as to ensure that the selected model can adapt to a wide range of usage scenarios. Next, the multiple far-field voice wake-up models in the model library are tested using the wake-up rate simulation tool. This process involves applying each model to the acquired audio samples and recording the number of successful wake-ups of each model to calculate its wake-up rate. In this way, the performance of each model in actual application can be objectively evaluated without having to deploy them one by one on the hardware for testing, which greatly reduces time and cost. Subsequently, based on the above wake-up rate data, a far-field voice wake-up model with the best wake-up effect is analyzed and selected as the target model. The key to this step is not only to select a model with a high wake-up rate, but also to consider the versatility and adaptability of the model to ensure that it can work well on different types of TV devices and improve the model reuse rate. Then, for the selected target far-field voice wake-up model, a corresponding APK file is generated, and the model resources are compiled into an application package (APK) that can run on the TV operating system. This step simplifies the software deployment process, so that the voice interaction function can be more conveniently installed on the TV terminal. Finally, the TV receives the target APK file of the target far-field voice wake-up model, and completes the installation of the target APK file to realize the voice interaction function. Since the entire adaptation process no longer requires frequent adjustment of the physical structure or acoustic characteristics, and no longer relies on on-site debugging by a third party, the development cycle is greatly shortened, the adaptation cost is reduced, and the efficiency and model compatibility are improved. In summary, through the method provided by the present invention, TV manufacturers can add far-field voice interaction functions to new products more quickly and economically without sacrificing quality, and ultimately provide consumers with a more intelligent and convenient user experience. This method not only optimizes the product development process, but also promotes the further development of the smart home ecosystem.

[0042] This embodiment provides a method for automatically adapting a TV far-field voice wake-up model, which can be applied to smart terminals, such as Figure 1 As shown in , the specific steps include:

[0043] Step S100: Acquire audio, and use a wake-up rate simulation tool to determine the wake-up rate of each far-field voice wake-up model for the audio, and obtain wake-up rate data.

[0044] In this embodiment, if Figure 2As shown, first, audio is obtained to provide a standardized and repeatable audio data set for subsequent model evaluation, ensuring that different far-field voice wake-up models are compared under the same conditions. The audio includes but is not limited to 120 wake-up words. The audio file can contain multiple wake-up words or instructions. The number and content of these wake-up words can be designed according to actual needs. Then, the wake-up rate simulation tool is used to determine the wake-up rate of each far-field voice wake-up model for the audio, and the wake-up rate data is obtained, allowing developers to quickly and accurately evaluate the performance of multiple models without relying on physical devices, greatly improving the screening efficiency and accuracy, while reducing the high time cost and resource waste caused by physical debugging.

[0045] Specifically, the step S100 includes the following steps:

[0046] Step S101: pre-processing the audio using a wake-up rate simulation tool to obtain processed audio, wherein the pre-processing includes any one or more of echo cancellation, noise reduction, dereverberation, and gain adjustment;

[0047] Step S102: Based on the processed audio, a wake-up rate simulation tool is used to determine the wake-up rate of each far-field voice wake-up model to the processed audio to obtain wake-up rate data.

[0048] In one implementation, the process of obtaining audio includes: first, using a USB to TTL cable to connect the computer and the tested TV, and starting the test tool to select the corresponding serial port input, ensuring stable communication between the computer and the TV, so that the tester can accurately control and monitor the reaction of the TV; then, the test scene can be determined in the test tool, and then based on the test scene, the audio corresponding to the test scene can be determined, so that the corresponding audio file can be accurately selected according to different test requirements to ensure the pertinence and effectiveness of the test. After the selection is completed, the display interface of the test tool displays the information of the test scene, providing intuitive operation guidance for the tester, facilitating the distinction and management of different test items, and improving work efficiency. Then, the computer is connected to the external speaker, and according to the requirements of different test scenes, the external speaker is placed at a preset position away from the TV, and the sound direction of the external speaker is opposite to the microphone hole of the TV, that is, the sound surface of the external speaker is facing the microphone hole of the TV, ensuring that the sound axis of the external speaker is aligned with the receiving axis of the TV microphone. Such an arrangement can simulate different listening conditions in the actual use environment, thereby obtaining a test result that is closer to the user's real experience. Next, pink noise was played through an external speaker and the sound pressure level of the external speaker under this condition was calibrated using a sound level meter to ensure the consistency of sound output under all test conditions. That is, no matter where the external speaker is located, the sound intensity it emits can be maintained at a standard level, providing a fair and reliable benchmark for subsequent audio playback tests in different scenarios, ensuring the comparability and accuracy of the test data. Finally, the external speaker plays the audio, and the TV records the audio through its microphone throughout the process to obtain the audio.

[0049] In one implementation, after obtaining the audio, based on the audio, the wake-up rate simulation tool inside the TV system is used to evaluate each far-field voice wake-up model, and the existing far-field voice wake-up models are switched one by one to complete the evaluation of all far-field voice wake-up models in the model library. Among them, the wake-up rate simulation tool contains two partial models, both of which are binary files (.bin format): one is the front-end audio signal processing, and the other is the scoring judgment. When the audio is imported into the tool, first, the audio is pre-processed using the wake-up rate simulation tool to obtain processed audio, wherein the pre-processing includes any one or more of echo cancellation, noise reduction, dereverberation, and gain adjustment. These pre-processing measures can significantly improve the audio quality, making subsequent scoring more accurate and reliable. For example, in a typical living room environment, the sound played by the TV may echo with the user's voice command. Through echo cancellation technology, this interference can be effectively reduced, allowing the far-field voice wake-up model to recognize voice commands more clearly. Noise reduction and dereverberation help to remove the influence of background noise and room reflections, while gain adjustment ensures that all test audios have similar volume levels, thus providing a fair competitive environment for each far-field voice wake-up model. Next, based on the processed audio, the wake-up rate simulation tool is used to determine the score of each far-field voice wake-up model for the processed audio. Assuming that there are three different far-field voice wake-up models: Model A, Model B, and Model C, the processed audio is input into the wake-up rate simulation tool. The wake-up rate simulation tool simulates the actual usage scenario and gives a score based on each model's recognition ability of the wake-up word. The score can be a probability value, indicating the possibility that the model believes that the audio contains the wake-up word. For example: for Model A, it scores 0.85 for the processed audio; for Model B, the score is 0.76; for Model C, the score is 0.92. The higher the score, the more accurate the response of the far-field voice wake-up model to the audio. For example, if a piece of audio contains the instruction "turn on the TV", the far-field voice wake-up model that can correctly recognize and execute this command will receive a higher score. Subsequently, based on the score and the preset threshold, the number of successful wake-ups for each far-field voice wake-up model whose score is equal to or greater than the preset threshold is determined. Assuming that the preset threshold is set to 0.8, then based on this threshold, it can be determined which scores represent successful wake-ups: Model A has a score of 0.85, which is greater than the preset threshold of 0.8, so it is counted as a successful wake-up; Model B has a score of 0.76, which is lower than the preset threshold of 0.8, so it is not counted as a successful wake-up; Model C has a score of 0.92, which is greater than the preset threshold of 0.8, so it is also counted as a successful wake-up. If this process is repeated multiple times (for example, using multiple different processed audios), the number of successful wake-ups for each model can be accumulated.The preset threshold is adjustable, which means that the tester can flexibly set the qualification standard according to actual needs. The purpose of this step is to quantify the performance of each model so as to clarify which far-field voice wake-up models perform well under specific conditions. Then, based on the number of successful wake-ups and the total number of scores of each far-field voice wake-up model, the wake-up rate of each far-field voice wake-up model for the processed audio is determined to obtain the wake-up rate data. The wake-up rate can be obtained through the formula. The wake-up rate data intuitively reflects the effectiveness of the far-field voice wake-up model in processing audio, and provides a scientific basis for selecting the best far-field voice wake-up model. Assuming that a total of 100 tests were conducted: Model A successfully woke up 85 times, Model B successfully woke up 60 times, and Model C successfully woke up 90 times, then the wake-up rate data are as follows: The wake-up rate of Model A is The wake-up rate of Model B is Or 60%, the wake-up rate of Model C is For example, if Model C successfully wakes up 90 times out of 100 attempts, its wake-up rate is 90%, indicating that the model is relatively reliable in this type of audio environment. In addition, when the wake-up rate simulated by a far-field voice wake-up model for audio does not meet the standard, that is, it is lower than the preset wake-up rate, the tester can flexibly adjust the preset wake-up rate according to actual needs, and the TV will automatically switch to other far-field voice wake-up models and re-simulate the wake-up rate for the same audio until all far-field voice wake-up models in the model library are switched. This mechanism ensures that even if the far-field voice wake-up model initially selected is not suitable, the best far-field voice wake-up model that best suits the current audio environment can always be found, thereby improving the robustness of the overall system and user experience. In summary, a series of steps from connecting the test equipment, selecting and calibrating the test scene, to obtaining audio and using the wake-up rate simulation tool inside the TV to evaluate the far-field voice wake-up model ensure the authenticity of the test environment and the accuracy of the test results.

[0050] Step S200: obtaining a target far-field voice wake-up model based on the wake-up rate data.

[0051] In one implementation, Figure 2As shown, first, the wake-up rate data is sorted by recursive sorting using a wake-up rate simulation tool. Recursive sorting is an efficient sorting algorithm that can ensure that the sorting work can be completed quickly and accurately even when faced with a large amount of wake-up rate data. The advantage of this sorting method is that it can not only process a small amount of data sets, but also efficiently cope with large-scale data sets, ensuring the accuracy and reliability of the sorting results. For example, in a test environment with dozens of different far-field voice wake-up models, recursive sorting can arrange these far-field voice wake-up models from high to low according to their wake-up rates in a very short time, providing a solid foundation for subsequent screening. Then, through the sorted data, the far-field voice wake-up model with the highest wake-up rate can be screened out to obtain the target far-field voice wake-up model. The purpose of this process is to identify and select the far-field voice wake-up model with the best performance, so as to ensure that the far-field voice wake-up model can provide the most stable and reliable voice wake-up service in practical applications. For example, if after recursive sorting, it is found that the wake-up rate of a certain far-field voice wake-up model in multiple tests is significantly higher than that of other far-field voice wake-up models (for example, a success rate of 98%), then this far-field voice wake-up model will be selected as the target far-field voice wake-up model because it performs well and can maintain a high wake-up rate in various complex audio environments, which is crucial to improving the user experience. In summary, by recursively sorting the wake-up rate data and screening out the target far-field voice wake-up model with the highest wake-up rate, not only the test efficiency is improved, but also the final selected target far-field voice wake-up model has the best performance. This method not only helps to improve the robustness and response speed of the system, but also provides users with a more intelligent and humanized voice interaction experience.

[0052] Step S300: Receive the target APK file of the target far-field voice wake-up model, and complete the installation of the target APK file.

[0053] In one implementation, Figure 2As shown, the target far-field voice wake-up model should be compiled into an APK file and installed. In the process of integrating the target far-field voice wake-up model into the TV system, first, it is necessary to confirm the specific storage path of the target far-field voice wake-up model file in the TV system, where the target far-field voice wake-up model is a binary file (.bin format), because the correct storage path ensures that the application can accurately find and load the binary model file (.bin format). Usually, this information can be obtained by consulting the device documentation or communicating with the hardware developer. For example, in some smart TVs, a specific external storage area may be reserved for storing machine learning models, while other devices may require the model to be stored in a specific directory of the internal storage. Then, based on the specific storage path, write the source code of the application of the target far-field voice wake-up model, which includes writing code in Java or Kotlin to implement the function of loading and initializing the target far-field voice wake-up model from the specified path. In this way, developers can ensure that no matter where the model is stored, the application can reliably access and use it. For example, if the target far-field voice wake-up model is located at / storage / emulated / 0 / models / voice_wakeup_model.bin, the developer will specify this path in the code and write the corresponding logic to read and load the model. Subsequently, based on a software development kit (SDK), such as the Android SDK, the source code of the application of the target far-field voice wake-up model is compiled into a bytecode file, and the bytecode file and resource files are packaged into an APK file. For example, using tools in the Android SDK, such as using a javac or kotlinc compiler to convert the written Java or Kotlin code into bytecode files (.class files), these bytecode files contain application logic, define how to interact with the binary model, and use the aapt (Android Asset Packaging Tool) tool to package all necessary resource files (such as pictures, layout files, and binary model files) and compiled bytecode files into an APK file. This step ensures that all components are properly integrated to form a distributable application package. Then, based on the software development kit (SDK), the APK file is aligned and optimized, and the APK file is signed to obtain the target APK file. For example, the zipalign tool is used to align the APK file to ensure that the resource data in the APK file is aligned according to the 4-byte boundary, thereby improving the running efficiency of the application on the device, reducing memory usage, and speeding up the startup speed.For example, an unaligned APK may cause additional I / O operations to read unaligned data blocks when the application is loaded. After zipalign optimization, this unnecessary overhead is eliminated; then, the APK file after alignment optimization is signed by the jarsigner or apksigner tool to obtain the target APK file. The signing operation verifies the source of the APK file and ensures that the file has not been tampered with since it was signed. Finally, the TV receives the target APK file of the target far-field voice wake-up model and completes the installation of the target APK file. At this stage, the TV operating system parses the APK file, extracts and installs all necessary components therein, including the binary file of the target far-field voice wake-up model and its related dependencies. Once the installation is complete, the target far-field voice wake-up model can be put into use immediately and start responding to the user's far-field voice commands. In summary, the above process covers the complete chain from confirming the storage path, writing source code, compiling and packaging, alignment optimization, signing to final installation. Through the above steps, it is ensured that the target far-field voice wake-up model can run stably, safely and efficiently in the TV system, providing users with a more intelligent and reliable voice interaction experience.

[0054] In summary, this embodiment first obtains audio to provide a standardized and repeatable audio data set for subsequent model evaluation, ensuring that different far-field voice wake-up models are compared under the same conditions. Then, the wake-up rate simulation tool is used to determine the wake-up rate of each far-field voice wake-up model for the audio, and the wake-up rate data is obtained, allowing developers to quickly and accurately evaluate the performance of multiple models without relying on physical devices, greatly improving the screening efficiency and accuracy, while reducing the high time cost and resource waste caused by physical debugging. Subsequently, based on the wake-up rate data, the target far-field voice wake-up model is obtained, which improves the scientificity and accuracy of model selection, enhances the possibility of model reuse, and reduces the development cost and time required for adapting new models. Then, the target APK file of the target far-field voice wake-up model is received, and the installation of the target APK file is completed, realizing seamless connection from model selection to application deployment, simplifying the development process, and improving work efficiency. The method proposed in the present invention effectively solves the problems of frequent third-party intervention, high time cost required for machine delivery back and forth, and low efficiency of single model adaptation in the adaptation process of traditional voice interaction solutions by introducing simulation screening technology. The proposed method not only optimizes the far-field voice interaction performance and improves the reuse rate of the model, but also significantly shortens the adaptation cycle and reduces the development cost. Overall, this method greatly improves the development efficiency and quality of voice interaction functions, and provides a more efficient and convenient voice interaction solution for devices such as smart TVs.

[0055] like Figure 3As shown in , this embodiment also provides a system for automatically adapting a TV far-field voice wake-up model, which includes: a wake-up rate data acquisition module 10, a target far-field voice wake-up model acquisition module 20, and an installation module 30. Specifically, the wake-up rate data acquisition module 10 is used to acquire audio, and use a wake-up rate simulation tool to determine the wake-up rate of each far-field voice wake-up model for the audio to obtain wake-up rate data. The target far-field voice wake-up model acquisition module 20 is used to obtain a target far-field voice wake-up model based on the wake-up rate data. The installation module 30 is used to receive a target APK file of the target far-field voice wake-up model and complete the installation of the target APK file.

[0056] In one implementation, obtaining audio includes:

[0057] Acquire a test tool, and determine a test scenario based on the test tool;

[0058] Based on the test scenario, the audio corresponding to the test scenario is obtained.

[0059] In one implementation, the wake-up rate data acquisition module 10 includes:

[0060] A processed audio acquisition unit, used to pre-process the audio using a wake-up rate simulation tool to obtain processed audio, wherein the pre-processing includes any one or more of echo cancellation, noise reduction, dereverberation, and gain adjustment;

[0061] The wake-up rate data acquisition unit determines the wake-up rate of each far-field voice wake-up model to the processed audio by using a wake-up rate simulation tool based on the processed audio to obtain the wake-up rate data.

[0062] In one implementation, the wake-up rate data acquisition unit includes:

[0063] a score determination subunit, configured to determine, based on the processed audio, a score of the processed audio by each far-field speech wake-up model using a wake-up rate simulation tool;

[0064] The first subunit for acquiring wake-up rate data is used to determine the wake-up rate of each far-field voice wake-up model for the processed audio based on the score to obtain the wake-up rate data.

[0065] In one implementation, the wake-up rate data acquisition first subunit includes:

[0066] A successful wake-up number determination subunit, used to determine the number of successful wake-up times for each far-field voice wake-up model whose score is equal to or greater than the preset threshold based on the score and the preset threshold;

[0067] The second subunit for acquiring wake-up rate data is used to determine the wake-up rate of each far-field voice wake-up model for the processed audio based on the number of successful wake-ups and the total number of scoring times of each far-field voice wake-up model, and obtain the wake-up rate data.

[0068] In one implementation, the target far-field voice wake-up model acquisition module 20 includes:

[0069] The target far-field voice wake-up model acquisition unit is used to use the wake-up rate simulation tool to sort the wake-up rate data through recursive sorting, screen out the far-field voice wake-up model with the highest wake-up rate, and obtain the target far-field voice wake-up model.

[0070] In one implementation, the receiving of the target APK file of the target far-field voice wake-up model and completing the installation of the target APK file includes:

[0071] An APK file acquisition unit, used to compile the source code of the application of the target far-field voice wake-up model into a bytecode file based on a software development kit, and package the bytecode file and resource files into an APK file;

[0072] The target APK file acquisition unit is used to perform alignment optimization on the APK file based on a software development kit, and to sign the APK file to obtain a target APK file.

[0073] The working principles of each module in the automatic adaptation TV far-field voice wake-up model system of this embodiment are the same as the principles of each step in the above method embodiment, and will not be repeated here.

[0074] Based on the above embodiment, the present invention further provides a terminal, the principle block diagram of the terminal can be as follows: Figure 4 The terminal may include one or more processors 100 ( Figure 4 Only one is shown), a memory 101 and a computer program 102 stored in the memory 101 and executable on one or more processors 100, for example, a program for automatically adapting a TV far-field voice wake-up model. When one or more processors 100 execute the computer program 102, the various steps in the embodiment of the method for automatically adapting a TV far-field voice wake-up model can be implemented. Alternatively, when one or more processors 100 execute the computer program 102, the functions of the various modules / units in the embodiment of the method for automatically adapting a TV far-field voice wake-up model can be implemented, which is not limited here.

[0075] In one embodiment, the processor 100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0076] In one embodiment, the memory 101 may be an internal storage unit of an electronic device, such as a hard disk or memory of the electronic device. The memory 101 may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device. Further, the memory 101 may also include both an internal storage unit of the electronic device and an external storage device. The memory 101 is used to store computer programs and other programs and data required by the terminal. The memory 101 may also be used to temporarily store data that has been output or is to be output.

[0077] Those skilled in the art will understand that Figure 4 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0078] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, operating database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double operational data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically adapting a TV far-field voice wake-up model, characterized in that: The method comprises: Acquire audio, and use a wake-up rate simulation tool to determine the wake-up rate of each far-field voice wake-up model for the audio, to obtain wake-up rate data; Based on the wake-up rate data, a target far-field voice wake-up model is obtained; Receive the target APK file of the target far-field voice wake-up model, and complete the installation of the target APK file.

2. The method for automatically adapting a TV far-field voice wake-up model according to claim 1, characterized in that: The step of obtaining audio includes: Acquire a test tool, and determine a test scenario based on the test tool; Based on the test scenario, the audio corresponding to the test scenario is obtained.

3. The method for automatically adapting a TV far-field voice wake-up model according to claim 2, characterized in that: The using of the wake-up rate simulation tool to determine the wake-up rate of each far-field voice wake-up model to the audio, and obtaining the wake-up rate data, includes: Using a wake-up rate simulation tool, preprocessing the audio to obtain processed audio, wherein the preprocessing includes any one or more of echo cancellation, noise reduction, dereverberation, and gain adjustment; Based on the processed audio, a wake-up rate simulation tool is used to determine the wake-up rate of each far-field voice wake-up model to the processed audio to obtain wake-up rate data.

4. The method for automatically adapting a TV far-field voice wake-up model according to claim 3, characterized in that: The method of determining the wake-up rate of each far-field voice wake-up model to the processed audio using a wake-up rate simulation tool based on the processed audio to obtain wake-up rate data includes: Based on the processed audio, using a wake-up rate simulation tool to determine a score of each far-field speech wake-up model on the processed audio; Based on the score, the wake-up rate of each far-field voice wake-up model for the processed audio is determined to obtain wake-up rate data.

5. The method for automatically adapting a TV far-field voice wake-up model according to claim 4, characterized in that: The step of determining the wake-up rate of each far-field voice wake-up model for the processed audio based on the score to obtain wake-up rate data includes: Based on the score and the preset threshold, determining the number of successful wake-up times for each far-field voice wake-up model whose score is equal to or greater than the preset threshold; Based on the number of successful wake-up times and the total number of scoring times of each far-field voice wake-up model, the wake-up rate of each far-field voice wake-up model for the processed audio is determined to obtain wake-up rate data.

6. The method for automatically adapting a TV far-field voice wake-up model according to claim 5, characterized in that: The obtaining a target far-field voice wake-up model based on the wake-up rate data includes: The wake-up rate simulation tool is used to sort the wake-up rate data by recursive sorting, and the far-field voice wake-up model with the highest wake-up rate is screened out to obtain the target far-field voice wake-up model.

7. The method for automatically adapting a TV far-field voice wake-up model according to claim 6, characterized in that: The receiving the target APK file of the target far-field voice wake-up model and completing the installation of the target APK file includes: Based on a software development kit, compile the source code of the application of the target far-field voice wake-up model into a bytecode file, and package the bytecode file and the resource file into an APK file; Based on the software development kit, the APK file is aligned and optimized, and the APK file is signed to obtain a target APK file.

8. A system for automatically adapting a TV far-field voice wake-up model, characterized in that: The system comprises: A wake-up rate data acquisition module is used to acquire audio, and use a wake-up rate simulation tool to determine the wake-up rate of each far-field voice wake-up model for the audio, and obtain wake-up rate data; A target far-field voice wake-up model acquisition module, used to obtain a target far-field voice wake-up model based on the wake-up rate data; An installation module is used to receive the target APK file of the target far-field voice wake-up model and complete the installation of the target APK file.

9. A terminal, characterized in that: The terminal includes a memory, a processor, and an automatic adaptation TV far-field voice wake-up model program stored in the memory and runable on the processor. When the processor executes the automatic adaptation TV far-field voice wake-up model program, the steps of the automatic adaptation TV far-field voice wake-up model method as described in any one of claims 1-7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program for automatically adapting the TV far-field voice wake-up model. When the program for automatically adapting the TV far-field voice wake-up model is executed by the processor, the steps of the method for automatically adapting the TV far-field voice wake-up model as described in any one of claims 1-7 are implemented.

Images