Voice Recognition Keyword Library Segmentation for Mobile Terminals

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.

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.

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.

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.

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.

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.

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.

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.