Language Model Biasing Modulation for ASR Context Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated speech recognition (ASR) systems face challenges in reducing transcription word error rates due to biases in language models that are not adequately adjusted based on context changes.
Innovation Solution
The system modulates language model biasing by using context data such as time intervals between user queries and responses, gaze tracking data, and application switching data to calculate a context confidence score, which determines whether to adjust and interpolate language model biasing parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If language model biasing is applied to improve transcription accuracy, then word error rates decrease, but the system may incorrectly bias transcriptions when context changes occur
Solution Approach 1:
The system dynamically adjusts language model biasing parameters based on real-time context confidence scores. When context changes are detected (e.g., user switches topics or applications), the biasing parameters are modulated to reduce or eliminate previous context biases, allowing the system to adapt to new contexts while maintaining accurate transcriptions.
Solution Approach 2:
The system changes the parameters of the language model by interpolating between different biasing states based on context confidence scores. When confidence in the current context is low, the system interpolates parameters to reduce biasing strength, effectively transitioning the model from a biased state to a less biased or unbiased state appropriate for the new context.
2Measurement precision
If the system uses multiple context data sources to improve context detection accuracy, then context confidence scores become more reliable, but system complexity increases
Solution Approach 1:
The system merges multiple context data sources including gaze tracking data, application switching data, and temporal information into a unified context confidence score. By combining these diverse data sources, the system achieves more reliable context detection without requiring separate processing pipelines for each data type, as the fusion occurs at the confidence scoring level.
Solution Approach 2:
The context confidence score mechanism serves multiple functions: it evaluates context relevance, determines biasing parameter modulation, and adapts to various context change scenarios (topic switches, application changes, user inactivity). This multi-functional approach reduces overall system complexity by using a single mechanism for diverse context adaptation needs.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for modulating language model biasing. In some implementations, context data is received. A likely context associated with a user is determined based on at least a portion of the context data. One or more language model biasing parameters based at least on the likely context associated with the user is selected. A context confidence score associated with the likely context based on at least a portion of the context data is determined. One or more language model biasing parameters based at least on the context confidence score is adjusted. A baseline language model based at least on the one or more of the adjusted language model biasing parameters is biased. The baseline language model is provided for use by an automated speech recognizer (ASR).