Dynamic Grammar Support in WFST-Based ASR Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
WFST-based ASR systems suffer from poor recognition accuracy due to their inability to support dynamic grammars, limiting language changes, new word introduction, and personalization, which impairs user experience and utility.
Innovation Solution
The system dynamically composes a user-specific WFST by replacing non-terminals with sub-grammars derived from user-specific usage data, incorporating context-dependent and lexicon transducers, and using weighted phone loops to generate mono-phone words, allowing for on-the-fly modification of recognition cascades during runtime.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If static grammars are used in WFST-based ASR systems, then system simplicity and stability are maintained, but recognition accuracy and adaptability deteriorate due to inability to support language changes, new words, and personalization
Solution Approach 1:
The patent implements dynamic grammar support by allowing the WFST system to modify its grammar structures during runtime based on user-specific data. The system transitions from static grammars to dynamic grammars that can adapt to language changes, new word introduction, and personalization needs while maintaining system stability through controlled modification mechanisms.
Solution Approach 2:
The system changes the parameters of the grammar structures by incorporating user-specific data such as contact lists, application usage patterns, and media interaction frequencies. These parameter changes enable the grammar to adapt to individual user behaviors and preferences, improving recognition accuracy without compromising system stability.
2Adaptability or versatility
If dynamic grammars are supported in WFST-based ASR systems, then recognition accuracy and adaptability improve, but system complexity increases due to on-the-fly modification requirements
Solution Approach 1:
The patent segments the grammar system into modular components that can be independently modified and composed. By dividing the complex grammar structure into manageable segments, the system can perform on-the-fly modifications without overwhelming complexity, allowing dynamic adaptability while maintaining manageable system architecture.
Solution Approach 2:
The system performs preliminary actions by pre-processing and organizing user-specific data before incorporating it into the grammar structures. This preparation work reduces the complexity of runtime modifications, as the system only needs to integrate pre-processed data rather than handling raw information during speech recognition operations.
3Measurement precision
If user-specific data is incorporated into WFST, then personalization and recognition accuracy improve, but data processing requirements and system resource usage increase
Solution Approach 1:
The patent applies partial action by selectively incorporating only the most relevant user-specific data into the grammar structures based on usage frequencies and importance weights. Rather than processing all available user data equally, the system focuses on high-impact data points, reducing processing requirements while maintaining high recognition accuracy for personalized content.
Data Source
AI summary
Systems and processes are disclosed for recognizing speech using a weighted finite state transducer (WFST) approach. Dynamic grammars can be supported by constructing the final recognition cascade during runtime using difference grammars. In a first grammar, non-terminals can be replaced with a, weighted phone loop that produces sequences of mono-phone words. In a second grammar, at runtime, non-terminals can be replaced with sub-grammars derived from user-specific usage data including contact, media, and application lists. Interaction frequencies associated with these entities can be used to weight certain words over others. With all non-terminals replaced, a static recognition cascade with the first grammar can be composed with the personalized second grammar to produce a user-specific WEST. User speech can then be processed to generate candidate words having associated probabilities, and the likeliest result can be output.