Speech Recognition Co-Articulation Correction via Dynamic Acoustic Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Speech recognition systems face challenges in accurately identifying numeric sequences with co-articulation problems, particularly in tonal languages like Mandarin, where digit strings are often misrecognized due to co-articulation issues, leading to inaccuracies such as 'wu, wu' being misinterpreted as 'wu, er, wu' or 'er, er, er'.
Innovation Solution
The method involves using a dynamically generated possible numeric sequence and/or supplemental acoustic models to detect and correct co-articulation problems in digit strings, employing a base acoustic model and confusability matrices to identify and address common co-articulation issues, switching to supplemental models when necessary to improve recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a base acoustic model is used for speech recognition, then the system is simple and fast, but it misrecognizes numeric sequences with co-articulation problems (e.g., '5, 5, 5' becomes 'wu, wu', 'wu, er, wu', or 'er, er, er')
Solution Approach 1:
The system segments the speech recognition process into multiple stages: initial recognition with a base acoustic model, detection of co-articulation problems through confusability matrices, and correction using supplemental acoustic models. This segmentation allows the system to maintain simplicity for common cases while adding complexity only when needed to resolve co-articulation issues in numeric sequences.
Solution Approach 2:
The system performs preliminary action by pre-calculating confusability matrices that identify potential co-articulation problems before final recognition. This allows the system to prepare correction rules in advance, so when co-articulation is detected, the system can quickly switch to supplemental acoustic models without adding significant real-time complexity.
2Reliability
If supplemental acoustic models are added to handle co-articulation problems, then recognition accuracy improves, but the system complexity increases
Solution Approach 1:
The system applies local quality by using different acoustic models for different parts of the recognition process. The base acoustic model handles general speech recognition, while supplemental acoustic models are applied only locally to specific numeric sequences that exhibit co-articulation problems. This is determined through confusability matrices that identify which models are needed for which token sequences.
Solution Approach 2:
The system implements dynamics by making the acoustic model selection flexible and adaptive. Instead of using a single static model, the system dynamically switches between base and supplemental acoustic models based on real-time detection of co-articulation problems. This allows the system to optimize accuracy for difficult cases while maintaining simplicity for straightforward recognition tasks.
Data Source
AI summary
A method for recognizing speech involves reciting, into a speech recognition system, an utterance including a numeric sequence that contains a digit string including a plurality of tokens and detecting a co-articulation problem related to at least two potentially co-articulated tokens in the digit string. The numeric sequence may be identified using i) a dynamically generated possible numeric sequence that potentially corresponds with the numeric sequence, and/or ii) at least one supplemental acoustic model. Also disclosed herein is a system for accomplishing the same.