Speech Recognition Command Word Segmentation and Dynamic Update
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems face challenges in maintaining high recognition rates and efficient processing due to large word lists, noise interference, and the need for frequent updates of acoustic models when the word list or command set changes, which leads to increased calculation time and model size.
Innovation Solution
A method and device for updating pronunciation dictionaries and speech models online, allowing for dynamic addition of new command words and corresponding language models, reducing the need for offline updates and minimizing the number of frequently used words, thus enhancing recognition speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large word list is used in speech recognition, then the system can recognize more vocabulary, but the recognition rate decreases and calculation time increases
Solution Approach 1:
The patent segments the large vocabulary into two distinct parts: a command word list (small, frequently updated) and a general word list (large, stable). The speech recognition system first matches input against the small command word list, and only if no match is found, proceeds to the larger general word list. This segmentation allows high recognition accuracy for commands while maintaining broad vocabulary coverage.
2Adaptability or versatility
If the word list or command set is updated frequently, then the system remains current and adaptable, but the acoustic model must be retrained which increases calculation time and processing overhead
Solution Approach 1:
The patent extracts the command word list from the complete vocabulary and treats it as a separate, independent component. This extracted command subset can be updated without affecting the general word list or requiring full model retraining. The system maintains separate data structures for commands and general vocabulary, allowing independent updates of the command portion.
Solution Approach 2:
The patent implements dynamic updating mechanisms where the command word list can be modified in real-time based on user needs. The system dynamically adjusts the command vocabulary without requiring static pre-definition, allowing flexible addition, removal, or modification of command words while maintaining system performance.
3Measurement precision
If offline updates are performed to maintain accuracy, then recognition rate is preserved, but response time increases and system availability decreases
Solution Approach 1:
The patent performs preliminary organization of vocabulary into command and general sections during system initialization. This preliminary action establishes efficient data structures and indexing that enable rapid online updates without requiring full offline processing. The command word list is pre-configured for fast comparison operations.
4Adaptability or versatility
If a comprehensive word list is used, then the system handles diverse input, but the model file size increases
Solution Approach 1:
The patent segments the model into separate components: a compact command word section and a general vocabulary section. This segmentation allows the system to load only the essential command portion into memory for rapid processing, while the larger general word list can be stored externally or loaded selectively, reducing the active model footprint.
Data Source
AI summary
A speech recognition method and device are disclosed. The method includes: acquiring a text file specified by a user, and extracting a command word from the text file, to obtain a command word list; comparing the command word list with a command word library, to confirm whether the command word list includes a new command word; if the command word list includes the new command word, generating a corresponding new pronunciation dictionary; merging the new language model into a language model library; and receiving speech, and performing speech recognition on the speech according to an acoustic model, a phonation dictionary, and the language model library. Command words acquired online are closely related to online content; therefore, the number of the command words is limited and far less than the number of frequently used words.