Parse Tree Confidence Scoring for Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech recognition systems face challenges in accurately determining confidence levels, particularly in complex spoken language and varied acoustic environments, leading to ineffective and annoying dialogues when attempting to identify uncertain phrases within sentences.
Innovation Solution
A confidence computing system and method that computes confidence scores for all levels of a parse tree, using rich sets of acoustic, syntactic, and semantic features, allowing for precise identification of reliable and unreliable regions, thereby enabling more effective and user-friendly dialogues by limiting requests for repetition to specific phrases rather than entire sentences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional speech recognition systems calculate confidence scores only at word level or utterance level, then the system complexity remains manageable, but the precision of identifying unreliable phrases is insufficient
Solution Approach 1:
The patent segments the utterance into a parse tree structure with multiple levels (word level, phrase level, utterance level), allowing confidence scores to be calculated at each hierarchical level. This segmentation enables precise identification of unreliable phrases without requiring complete sentence repetition, as each node in the parse tree can be independently evaluated for confidence.
Solution Approach 2:
The patent introduces a hierarchical dimensional structure (parse tree with multiple levels) to organize confidence score calculation. By adding this dimensional hierarchy beyond traditional flat word-level or utterance-level approaches, the system achieves finer-grained precision in identifying unreliable regions while maintaining manageable complexity through structured organization.
2Reliability
If the system requests repetition of entire sentences when confidence is low, then it ensures reliability of processed information, but it increases loss of time and reduces user experience
Solution Approach 1:
The patent extracts and identifies only the specific unreliable phrase nodes from the parse tree that have low confidence scores, rather than treating the entire utterance as unreliable. This extraction approach allows the system to maintain reliability by focusing on problematic regions while minimizing time loss by requiring repetition only of the extracted unreliable phrases rather than entire sentences.
Solution Approach 2:
The patent applies local quality assessment by evaluating confidence scores at different regions (nodes) of the parse tree independently. This allows the system to identify that only specific local regions (phrases) may be unreliable while other regions remain reliable, enabling targeted repetition requests that preserve overall dialogue efficiency and reduce unnecessary user burden.
3Measurement precision
If the system uses rich sets of acoustic, syntactic, and semantic features for confidence calculation, then the measurement precision of confidence scores improves, but the computational complexity increases
Solution Approach 1:
The patent segments the computational task by calculating confidence scores at each hierarchical level of the parse tree separately (word level, phrase level, utterance level). This segmentation allows the rich feature sets (acoustic, syntactic, semantic) to be processed in a structured, modular manner, improving measurement precision through comprehensive feature analysis while managing computational complexity through hierarchical organization and incremental processing.
Data Source
AI summary
In a confidence computing method and system, a processor may interpret speech signals as a text string or directly receive a text string as input, generate a syntactical parse tree representing the interpreted string and including a plurality of sub-trees which each represents a corresponding section of the interpreted text string, determine for each sub-tree whether the sub-tree is accurate, obtain replacement speech signals for each sub-tree determined to be inaccurate, and provide output based on corresponding text string sections of at least one sub-tree determined to be accurate.