Voice Recognition Keyword Library Segmentation for Mobile Terminals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional voice recognition methods for mobile terminals suffer from low accuracy and high error rates, especially when processing multiple words or long sentences, leading to user dissatisfaction due to the need for repeated input and revisions.
Innovation Solution
A voice recognition method that classifies operation categories based on service functions and uses a keyword library specific to each category, allowing for efficient keyword extraction and matching, reducing the number of processed objects and improving recognition accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional voice recognition methods process multiple words or long sentences, then the system can handle more complex user instructions, but the recognition accuracy and speed significantly decrease
Solution Approach 1:
The patent segments the voice recognition process into two distinct phases: first identifying a trigger word to determine the operation category, then performing keyword recognition within that specific category. This segmentation allows the system to handle complex instructions by breaking them down into manageable parts, maintaining high accuracy in each phase while supporting versatile operations.
Solution Approach 2:
The patent applies local quality by using different recognition strategies for different parts of the voice input. The trigger word recognition uses one set of criteria, while the subsequent keyword recognition uses another set optimized for the specific operation category. This localized optimization ensures high accuracy for each component while maintaining overall system versatility.
2Adaptability or versatility
If conventional voice recognition methods process multiple words or long sentences, then the system can handle more complex user instructions, but the processing speed decreases
Solution Approach 1:
By segmenting the recognition process into trigger word identification followed by category-specific keyword recognition, the system processes complex instructions in stages rather than as a monolithic task. This segmentation enables parallel processing optimization and reduces the computational burden on each processing stage, maintaining high speed while handling versatile operations.
Solution Approach 2:
The trigger word recognition serves as a preliminary action that determines the operation category before the main keyword recognition takes place. This preliminary classification allows the system to prepare the appropriate keyword library and recognition parameters in advance, significantly speeding up the subsequent keyword matching process while still handling complex multi-word instructions.
3Adaptability or versatility
If the keyword library includes all possible operations, then the system can recognize any user instruction, but the search time and processing complexity increase
Solution Approach 1:
The patent segments the keyword library into multiple category-specific sub-libraries based on operation types (messaging, calling, music, etc.). Instead of searching through a single large library containing all possible operations, the system first identifies the operation category through trigger word recognition, then searches only within the relevant sub-library. This segmentation maintains comprehensive recognition coverage while dramatically reducing search time by limiting the search scope to the appropriate category.
Solution Approach 2:
The patent applies local quality by optimizing the keyword library structure for each operation category. Each category has its own tailored keyword library with terms and phrases specific to that operation type. This localized optimization ensures that the keyword matching process is both comprehensive for the specific category and efficient, as the search is confined to a smaller, more relevant subset of the overall keyword space.
4Ease of operation
If the system performs semantic analysis on natural language sentences, then the system can understand user intent, but the error rate in recognition and analysis increases
Solution Approach 1:
The patent segments the language processing task into identifying a trigger word (which indicates the operation category) followed by recognizing specific keywords within that category. This segmentation avoids the complexity of full semantic analysis on entire sentences, reducing error rates while still enabling natural language understanding through the structured approach of trigger word + keyword combinations.
Solution Approach 2:
The patent extracts the essential meaningful elements (trigger words and keywords) from natural language sentences, ignoring unnecessary words and grammatical structures. By focusing only on the critical keywords that indicate user intent within the context of the identified operation category, the system achieves reliable recognition while maintaining ease of operation with natural language input.
Data Source
AI summary
A voice recognition method and device, for improving efficiency and accuracy of voice recognition. The method comprises: receiving a trigger message of an operation class to be operated for operating on a mobile terminal, wherein the operation class is a class divided according to the service function of the mobile terminal (S101); receiving voice keyword information and determining a voice keyword from the voice keyword information (S102); and retrieving a keyword library under an operation class entry to be operated in accordance with the voice key word, and returning a search result (S103).

