Hybrid Text Generator for ASR Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated speech recognition (ASR) systems face challenges in achieving near-perfect recognition rates under suboptimal conditions such as background noise or thick accents, with no single system being dynamic and sophisticated enough to handle these variations effectively.
Innovation Solution
A hybrid text generator is developed to combine text strings from multiple ASR systems, using metadata like confidence scores and anchor words to create a more accurate hybrid text string by normalizing inputs, identifying anchor points, calculating match vectors, and resolving uncertainties through confidence score analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single ASR system is used, then device complexity is reduced, but recognition accuracy deteriorates under suboptimal conditions
Solution Approach 1:
The patent combines multiple ASR systems into a hybrid text generator that integrates their outputs. The system receives text strings from multiple ASR systems, aligns them using dynamic time warping, and merges them into a single hybrid text string by selecting words based on confidence scores and alignment positions, thereby achieving higher recognition accuracy than any single system alone.
Solution Approach 2:
The patent segments the text string generation process by dividing it into multiple independent ASR system outputs that are processed separately and then combined. Each ASR system independently processes the audio input and produces its own text string, which is then segmented into individual words for alignment and comparison in the hybrid text generation process.
2Measurement precision
If multiple ASR systems are combined, then recognition accuracy improves, but device complexity increases
Solution Approach 1:
The patent introduces an intermediary hybrid text generator that mediates between multiple ASR systems and the final output. This intermediary component aligns text strings using dynamic time warping, compares confidence scores, and synthesizes the hybrid text string, thereby managing the complexity of combining multiple systems while achieving improved accuracy.
Solution Approach 2:
The patent changes parameters by introducing confidence scores as a weighting mechanism and using dynamic time warping to adjust temporal alignments between text strings. By varying the alignment parameters and confidence thresholds, the system optimizes the combination of multiple ASR outputs to achieve superior recognition accuracy.
3Adaptability or versatility
If ASR systems use different algorithms, then adaptability to different audio conditions improves, but consistency of output deteriorates
Solution Approach 1:
The patent creates equipotentiality by normalizing the outputs of different ASR systems through dynamic time warping alignment. By adjusting the temporal parameters of each text string to a common reference frame, the system equalizes the positional correspondence of words across different ASR outputs, enabling consistent comparison and hybrid text generation despite algorithmic differences.
Data Source
AI summary
A hybrid text generator is disclosed that generates a hybrid text string from multiple text strings that are produced from an audio input by multiple automated speech recognition systems. The hybrid text generator receives metadata that describes a time-location that each word from the multiple text strings is located in the audio input. The hybrid text generator matches words between the multiple text strings using the metadata and generates a hybrid text string that includes the matched words. The hybrid text generator utilizes confidence scores associated with words that do not match between the multiple text strings to determine whether to add an unmatched word to the hybrid text string.