Dialog Manager Skip Lists for Speech Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Spoken computer-dialog systems often experience misunderstandings due to inaccurate automatic speech recognition, leading to repetitive errors and an annoying user experience, as the same mistakes can recur in subsequent dialog turns.
Innovation Solution
A dialog manager is configured to utilize skip lists, which contain improbable values for information items, to post-process or pre-process speech recognition results, allowing for the rejection and replacement of incorrect interpretations to improve accuracy and reduce misunderstandings in subsequent dialog turns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the ASR engine generates multiple alternative interpretations for each utterance, then the system has more options to choose from, but the complexity of processing and selecting the correct interpretation increases
Solution Approach 1:
The dialog manager pre-processes the speech recognition result by comparing alternative interpretations against the skip list before final selection. This preliminary filtering action removes improbable values early in the process, reducing the burden on subsequent processing stages and improving overall accuracy without significantly increasing complexity
Solution Approach 2:
The skip list acts as an intermediary data structure that mediates between the ASR engine's multiple interpretations and the dialog manager's selection process. It provides a pre-computed reference of improbable values that simplifies the filtering process, allowing the system to efficiently eliminate incorrect interpretations without complex real-time analysis
2Reliability
If the dialog manager uses confirmation questions to verify information, then misunderstanding can be detected, but the dialog becomes longer and more tedious
Solution Approach 1:
The system performs preliminary filtering of speech recognition results using the skip list before generating confirmation questions. By pre-identifying and removing improbable interpretations, the system reduces the likelihood of misunderstandings reaching the confirmation stage, thereby detecting fewer false positives and reducing unnecessary dialog extensions
Solution Approach 2:
The dialog manager extracts and removes improbable interpretations from the speech recognition result using the skip list, separating correct from incorrect interpretations before further processing. This extraction prevents incorrect values from triggering unnecessary confirmation questions, reducing dialog length while maintaining reliability
3Reliability
If the system prompts the user to repeat information after a misunderstanding, then the correct information can be obtained, but the same mistake is likely to occur again
Solution Approach 1:
The dialog manager implements feedback by maintaining a skip list that accumulates improbable values encountered during the dialog. When the same acoustic pattern is recognized again, the stored feedback in the skip list enables the system to immediately reject the incorrect interpretation, preventing repetitive misunderstandings and reducing the need for user repetition
Solution Approach 2:
The system performs preliminary filtering using the skip list before final interpretation selection, proactively preventing repetitive mistakes rather than reacting to them. By checking against previously encountered improbable values, the system eliminates the root cause of repetition before it occurs, reducing dialog turns and improving efficiency
Data Source
AI summary
A method for improving speech recognition accuracy includes utilizing skiplists or lists of values that cannot occur because of improbability or impossibility. A table or list is stored in a dialog manager module. The table includes a plurality of information items and a corresponding list of improbable values for each of the plurality of information items. A plurality of recognized ordered interpretations is received from an automatic speech recognition (ASR) engine. Each of the plurality of recognized ordered interpretations each includes a number of information items. A value of one or more of the received information items for a first recognized ordered interpretation is compared to a table to determine if the value of the one of the received information items matches any of the list of improbable values for the corresponding information item.