Correction Model for ASR Error Reduction in Dialog Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic speech recognition systems in dialog systems often propagate errors due to inaccuracies in converting spoken language to logical forms, leading to ineffective user interactions and task execution.
Innovation Solution
A correction model is implemented between the automatic speech recognition (ASR) and semantic parser subsystems, or integrated into the semantic parser, to correct ASR output by training on tuples of original and corrected utterances, reducing error propagation through the dialog system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If automatic speech recognition is used to convert spoken language to logical forms, then user interaction capability is improved, but error propagation occurs leading to reduced accuracy
Solution Approach 1:
A correction model is introduced as an intermediary component between the ASR subsystem and semantic parser subsystem. This correction model receives ASR output, identifies and corrects errors, and passes corrected output to the semantic parser, thereby mediating the error propagation problem while preserving the speech-based interaction capability
Solution Approach 2:
The system implements a feedback mechanism where the correction model analyzes ASR output and generates corrections based on learned error patterns. The correction process feeds back into the semantic parser input, creating a closed-loop system that continuously improves accuracy by correcting ASR errors before they propagate further
2Measurement precision
If a correction model is added between ASR and semantic parser subsystems, then error propagation is reduced, but device complexity increases
Solution Approach 1:
The correction model can be merged with the semantic parser subsystem, integrating error correction functionality within an existing component. This merging approach reduces the number of separate subsystems and interfaces, thereby lowering overall system complexity while maintaining the error correction benefit
Solution Approach 2:
The correction model is designed to handle multiple types of ASR errors universally, serving as a general-purpose correction layer that can address various error patterns without requiring separate specialized components for each error type, thus avoiding complexity multiplication
Data Source
AI summary
Some techniques described herein determine a correction model for a dialog system, such that the correction model corrects output from an automatic speech recognition (ASR) subsystem in the dialog system. A method described herein includes accessing training data. A first tuple of the training data includes an utterance, where the utterance is a textual representation of speech. The method further includes using an ASR subsystem of a dialog system to convert the utterance to an output utterance. The method further includes storing the output utterance in corrective training data that is based on the training data. The method further includes training a correction model based on the corrective training data, such that the correction model is configured to correct output from the ASR subsystem during operation of the dialog system.