Voice wake-up optimization method
By creating a non-standard activation dictionary and dynamically updating it based on user voice data, the problem of wake-up failure caused by fixed templates in voice wake-up technology has been solved, improving user experience and success rate.
Patent Information
- Application Number
- CN202411120883.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-08
AI Technical Summary
Existing voice wake-up technology suffers from wake-up failures due to fixed templates and differences in user accents, affecting user experience.
A non-standard activation word library is created. By collecting user voice data, wake words with clear activation intentions are identified and labeled. The non-standard activation word library is dynamically updated to match user habits, and an ASR model is used for identification and comparison.
It improves the success rate of voice wake-up, enhances the user experience, and the non-standard activation dictionary is more in line with user habits and offers greater flexibility.
Smart Images

Figure CN118918895A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of voice wake-up, and in particular to a voice wake-up optimization method. Background Art
[0002] With the development of voice interaction technology, more and more smart devices have begun to use voice interaction technology. When a device enters the voice interaction working state, it is necessary to perform a voice wake-up operation first. As the entrance to voice interaction, voice wake-up technology has become an indispensable function in various devices.
[0003] The existing voice wake-up technology first builds a template, and when encountering voice input, it matches the voice input with the template, and then identifies whether the voice input is a wake-up word. Due to the fixed template and problems such as the user's accent, it is easy to cause the voice input to not match the template, resulting in wake-up failure, which seriously affects the user experience. Summary of the invention
[0004] In order to improve the user experience, this application provides a voice wake-up optimization method.
[0005] The technical solution adopted by the present invention to solve the above problems is:
[0006] Voice wake-up optimization methods include:
[0007] Step 1: Create a non-standard activation word library, in which wake-up words that are inconsistent with standard wake-up words but have clear activation intentions are stored;
[0008] Step 2: Collect and recognize the voice data of the user waking up the terminal, and compare the recognition result with the standard wake-up word. If they are consistent, the voice data is the wake-up voice. If they are inconsistent, compare it with the non-standard activation word library. If the comparison with a non-standard activation word in the non-standard activation word library is successful, the voice data is the wake-up voice.
[0009] Furthermore, the steps for creating a non-standard activation vocabulary are:
[0010] Step 11: Collect the voice data of the user waking up the terminal;
[0011] Step 12: Recognize the voice data, and save the recognition results that are inconsistent with the standard wake-up word into an inconsistent folder;
[0012] Step 13, counting the types and corresponding quantities of recognition results in the inconsistent folders;
[0013] Step 14: annotate the statistically identified results with clear activation intentions, where the annotated information at least includes the activation intentions;
[0014] Step 15: If the number of the marked recognition results is greater than the threshold, the speech data is added to the non-standard activation vocabulary.
[0015] Furthermore, the step 2 also includes: if the recognition result fails to be compared with the non-standard activation word, the current recognition result is stored in an inconsistent folder.
[0016] Further, after the preset time arrives, step 13 to step 15 are executed.
[0017] Furthermore, the preset time is 1 month.
[0018] Furthermore, the ASR model is used to recognize the speech data.
[0019] Compared with the prior art, the present invention has the following beneficial effects: a non-standard activation word library is created, which contains wake-up words that are inconsistent with the standard wake-up words but have clear activation intentions. When the user's voice input does not match the standard template, it can be matched with the non-standard activation word to achieve wake-up, thereby improving the user experience. The non-standard activation word library is established based on the user's voice input, which is more in line with the user's usage habits; through regular dynamic updates, the non-standard activation word library is more flexible. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 Flowchart of the method for voice wake-up optimization. DETAILED DESCRIPTION
[0021] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. 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.
[0022] like Figure 1 As shown, the voice wake-up optimization method includes:
[0023] Step 1: Create a non-standard activation word library, in which wake-up words that are inconsistent with standard wake-up words but have clear activation intentions are stored;
[0024] Step 2: Collect and recognize the voice data of the user waking up the terminal. The ASR model can be used to recognize the voice data and compare the recognition result with the standard wake-up word. If they are consistent, the voice data is the wake-up voice. If they are inconsistent, they are compared with the non-standard activation word library. If the comparison with a non-standard activation word in the non-standard activation word library is successful, the voice data is the wake-up voice. After being determined to be the wake-up voice, the wake-up operation can be performed.
[0025] In this embodiment, the non-standard activation word library is generated according to the user's daily wake-up voice to better fit the user's usage habits. The specific steps are:
[0026] Step 11: Collect the voice data of the user waking up the terminal;
[0027] Step 12: Recognize the voice data, and store the recognition results that are inconsistent with the standard wake-up word into an inconsistent folder; for example, if the standard activation word is XX Xiaobai, the recognition results may be: XX xiao bai, XX xiu bei, XX xiao bei, etc.; then XX xiu bei and XX xiao bei are stored into an inconsistent folder;
[0028] Step 13: Count the types and corresponding quantities of the recognition results in the inconsistent folders; this embodiment adopts a regular statistical method, such as one month;
[0029] Step 14: manually label the statistically identified results with clear activation intentions, where the labeling information at least includes the activation intention, such as labeling XX xiu bei as XX Xiaobai;
[0030] Step 15: If the number of marked recognition results is greater than the threshold, the voice data is added to the non-standard activation word library. The specific value of the threshold can be set according to actual needs and is not limited here. When the user's next voice recognition result is XX xiu bei, the same wake-up function as XX Xiaobai can be achieved by successfully comparing it with the non-standard activation word library.
[0031] Furthermore, if the recognition result fails to be compared with the non-standard activation word, the current recognition result is stored in an inconsistent folder to increase the diversity of the number of samples in the inconsistent folder.
Claims
1. Voice wake-up optimization method, characterized in that: include: Step 1: Create a non-standard activation word library, in which wake-up words that are inconsistent with standard wake-up words but have clear activation intentions are stored; Step 2: Collect and recognize the voice data of the user waking up the terminal, and compare the recognition result with the standard wake-up word. If they are consistent, the voice data is the wake-up voice. If they are inconsistent, compare it with the non-standard activation word library. If the comparison with a non-standard activation word in the non-standard activation word library is successful, the voice data is the wake-up voice.
2. The voice wake-up optimization method according to claim 1, characterized in that: The steps to create a non-standard activation vocabulary are: Step 11: Collect the voice data of the user waking up the terminal; Step 12: Recognize the voice data, and save the recognition results that are inconsistent with the standard wake-up word into an inconsistent folder; Step 13, counting the types and corresponding quantities of recognition results in the inconsistent folders; Step 14: annotate the statistically identified results with clear activation intentions, where the annotated information at least includes the activation intentions; Step 15: If the number of the marked recognition results is greater than the threshold, the speech data is added to the non-standard activation vocabulary.
3. The voice wake-up optimization method according to claim 2, characterized in that: The step 2 also includes: if the recognition result fails to be compared with the non-standard activation word, the current recognition result is stored in an inconsistent folder.
4. The voice wake-up optimization method according to claim 2, characterized in that: After the preset time arrives, execute step 13 to step 15.
5. The voice wake-up optimization method according to claim 4, characterized in that: The preset time is 1 month.
6. The voice wake-up optimization method according to any one of claims 1 to 5, characterized in that: The ASR model is used to recognize speech data.
Citation Information
Cited By
Wake-up word recognition method based on eigenvector iterative optimization
CN121214933A
