Personalized Speech Model Update via Grammar Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Spoken language understanding systems face challenges in maintaining user-specific personalization data up-to-date, particularly with changes in user activities such as new music purchases, which can lead to suboptimal performance in speech recognition and natural language understanding if personalized models are not quickly updated.
Innovation Solution
A system that efficiently generates and updates personal spoken language understanding models by using a data harvesting service to store and prioritize updates, allowing for rapid modification of grammar specifications and generation of personal models, focusing on recent user actions as strong predictors of subsequent behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complete personal models are regenerated to incorporate new user data, then model accuracy is improved, but processing time and computational resources are significantly consumed
Solution Approach 1:
The patent segments the model update process into two distinct components: grammar specification updates and model generation. The grammar specification is updated incrementally to reflect new user data, while the full model regeneration is deferred or performed less frequently. This segmentation allows the system to maintain accuracy by updating the grammar layer quickly without necessarily triggering a complete model regeneration cycle, thus reducing processing time while preserving model precision.
Solution Approach 2:
The system performs preliminary updates to the grammar specification whenever new user data is detected, preparing the model structure in advance. This preliminary action ensures that when full model regeneration is eventually performed, the grammar component is already optimized and ready, reducing the overall time required for complete model updates while maintaining accuracy.
2Measurement precision
If complete personal models are regenerated to incorporate new user data, then model accuracy is improved, but computational resources are significantly consumed
Solution Approach 1:
The patent divides the computational workload into two separate operations: grammar specification updates and model generation. By segmenting these tasks, the system can update the grammar specification with minimal computational resources whenever new user data arrives, while deferring the resource-intensive model generation process. This segmentation allows the system to maintain model accuracy through grammar updates without proportionally increasing computational resource consumption at each data update event.
3Measurement precision
If frequent model updates are performed to reflect recent user activities, then personalization accuracy is improved, but system complexity increases
Solution Approach 1:
The patent segments the personalization system into distinct layers: a grammar specification layer that handles frequent, simple updates based on user activities, and a model generation layer that performs comprehensive processing less frequently. This segmentation allows the system to achieve frequent personalization updates with low complexity grammar modifications, while the more complex model generation process is optimized and performed only when necessary, thus maintaining personalization accuracy without proportionally increasing overall system complexity.
Data Source
AI summary
Features are disclosed for maintaining data that can be used to personalize spoken language understanding models, such as speech recognition or natural language understanding models. The personalization data can be used to update the models based on some or all of the data. The data may be obtained from various data sources, such as applications or services used by the user. Personalized spoken language understanding models may be generated or updated based on updates to the personalization data or some other portion of the stored personalization data. Generation of personalized spoken language understanding models may be prioritized such that the generation process accommodates multiple users.