Character Sequence Scoring for Speech Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Speech recognition systems face difficulties in accurately interpreting unconstrained sequences of characters, such as names and points of interest, leading to frequent misidentification and user frustration, especially in navigation systems.
Innovation Solution
A speech recognition method and system that processes spoken utterances by selecting and scoring known character sequences, using a weighting of individual characters to identify probable matches, with a primary speech processor and character post-processor to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech recognition systems use conventional methods to interpret unconstrained character sequences, then the system complexity remains low, but the recognition accuracy deteriorates leading to frequent misidentification
Solution Approach 1:
The patent segments the character recognition process into distinct stages: phoneme-level recognition by the speech processor, followed by character sequence assembly and scoring by the post-processor. This segmentation allows each component to specialize, improving overall accuracy without requiring complete system redesign.
Solution Approach 2:
The patent applies preliminary action by pre-computing and storing character weights and sequence scores in databases before runtime. During speech recognition, the system retrieves and applies these pre-computed values rather than calculating them in real-time, enhancing recognition accuracy while maintaining manageable system complexity through prepared lookup tables.
2Reliability
If speech recognition systems process unconstrained character sequences with basic methods, then the processing speed remains fast, but the reliability deteriorates causing user frustration
Solution Approach 1:
The system performs preliminary actions by pre-calculating character weights based on frequency analysis and storing them in databases. During runtime, these pre-computed weights are quickly retrieved and applied to score candidate sequences, ensuring reliable identification without significant processing delays.
Solution Approach 2:
The patent implements feedback through the scoring mechanism that evaluates multiple candidate character sequences simultaneously. The system computes scores for each candidate based on character weights and phoneme matches, then selects the highest-scoring sequence, providing a reliable decision-making process that reduces misidentification while maintaining efficiency.
Data Source
AI summary
A method of and a system for processing speech. A spoken utterance of a plurality of characters can be received. A plurality of known character sequences that potentially correspond to the spoken utterance can be selected. Each selected known character sequence can be scored based on, at least in part, a weighting of individual characters that comprise the known character sequence.