ASR Error Correction via Historical Utterance Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Speech processing systems face errors due to inaccuracies in automatic speech recognition (ASR), leading to undesirable user experiences, especially when user inputs are misinterpreted due to background noise or other factors, resulting in unresponsive actions.
Innovation Solution
The system employs a trained model that leverages historical user data to determine the likelihood of success for user inputs, generating an alternate utterance by processing ASR hypotheses and historical user inputs, thereby reducing errors by selecting the most likely intended action based on past successful interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If automatic speech recognition is used to interpret user inputs, then speech-based control is enabled, but misinterpretation errors occur due to background noise
Solution Approach 1:
The system implements feedback by analyzing ASR confidence scores and error patterns, then using this information to generate corrected alternate utterances. The confidence score feedback mechanism triggers correction when accuracy falls below a threshold, and the error pattern feedback from historical data guides the correction process to improve future recognition accuracy.
Solution Approach 2:
The patent introduces an intermediary correction mechanism that sits between ASR and NLU. When ASR confidence is low or errors are detected, this intermediary layer generates alternate corrected utterances using historical error patterns, effectively mediating the translation from spoken input to accurate command interpretation.
2Measurement precision
If ASR hypotheses are processed to generate alternate utterances, then interpretation accuracy improves, but system complexity increases
Solution Approach 1:
The system performs preliminary action by pre-processing ASR hypotheses to generate confidence scores and pre-analyzing historical error patterns before NLU processing. This preliminary analysis allows the system to identify when corrections are needed and prepare alternate utterances in advance, reducing the computational burden during real-time operation.
Solution Approach 2:
The patent applies parameter changes by dynamically adjusting the level of correction based on ASR confidence scores. When confidence is high, no correction is applied; when confidence is low or errors are detected, correction mechanisms are activated. This parameter-based approach optimizes processing complexity by applying corrections only when necessary.
3Reliability
If historical user data is leveraged to generate alternate utterances, then error rates decrease, but data processing requirements increase
Solution Approach 1:
The system extracts only the relevant error patterns from historical user data that are specific to ASR misinterpretations. Rather than processing all historical data, it selectively extracts and applies correction patterns that are directly relevant to the current ASR hypotheses, reducing the volume of data that needs to be processed while maintaining high reliability.
Data Source
AI summary
A system is provided for handling errors during automatic speech recognition by leveraging past inputs spoken by the user. The system may process a user input to determine an ASR hypothesis. The system may then determine an alternate representation of the user input based on the inputs provided by the user in the past, and whether the ASR hypothesis sufficiently matches one of the past inputs.