ATC Transcription Merge Split Interface
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech transcription systems in aircraft environments face inaccuracies due to improper segmentation of transcribed text, leading to erroneous message splits and unsplit segments, which affect the accuracy of call sign identification and critical data extraction, especially in fast-paced ATC-pilot communications.
Innovation Solution
A system and method that allow users to review and manually merge or split text segments displayed on a transcription device, with a processor configured to update the graphical user interface based on user input, and log user actions to update the Automatic Speech Recognition model, enabling improved segmentation and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated speech recognition is used to transcribe ATC-pilot communications, then transcription speed is improved, but segmentation accuracy deteriorates due to variable speaking rates and pauses
Solution Approach 1:
The system displays transcribed text segments to users and receives feedback through manual merge or split actions. This feedback loop allows the system to learn from user corrections and improve segmentation accuracy over time, resolving the contradiction between automated speed and accuracy by incorporating human validation into the process.
Solution Approach 2:
The system enables users to directly correct segmentation errors through intuitive merge and split operations on displayed text segments. This self-service approach allows end-users to fix accuracy issues without requiring system reconfiguration, maintaining high transcription speed while enabling accuracy improvements through user intervention.
2Productivity
If text segments are automatically split based on probabilistic models, then processing efficiency is improved, but message integrity deteriorates due to erroneous splits and unsplit segments
Solution Approach 1:
The system performs preliminary automated segmentation to maintain processing efficiency, then presents the segmented text to users for review before final output. This preliminary action approach allows the system to benefit from automated processing while giving users the opportunity to correct errors, thus maintaining both efficiency and message integrity.
Solution Approach 2:
User corrections of split errors are captured and used to refine the probabilistic segmentation model. This feedback mechanism allows the system to learn from errors and improve message integrity over time while maintaining automated processing efficiency.
3Measurement precision
If manual review of transcribed messages is implemented, then segmentation accuracy is improved, but operational time increases
Solution Approach 1:
The system implements partial manual review by allowing users to review and correct only the segmentation portions they are uncertain about, rather than requiring complete manual review of all transcribed messages. This approach improves segmentation accuracy for problematic segments while minimizing the time loss associated with full manual review.
Solution Approach 2:
Users can quickly perform merge and split operations on displayed segments without extensive training or time investment. The self-service nature of the correction interface allows for rapid accuracy improvements without significant operational time penalties.
4Device complexity
If probabilistic segmentation models are used, then system complexity is reduced, but transcription quality deteriorates due to improper message splitting
Solution Approach 1:
User corrections of segmentation errors provide feedback that is used to refine the probabilistic model parameters. This feedback loop allows the simple probabilistic model to improve its transcription quality over time without increasing system complexity, as the model adapts to specific usage patterns and speaker characteristics.
Solution Approach 2:
The system automatically applies user corrections to improve future segmentation without requiring complex manual configuration or system reprogramming. The probabilistic model self-adjusts based on user feedback, maintaining low system complexity while improving transcription quality.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system is provided that includes a processor configured to receive an input of a user in response to a display of a text segment generated by a decode operation of an audio segment wherein the processor is configured to if the user determines by viewing a display containing one or more text segments that the one or more text segments include an improper splitting of text segments, perform a merge action in response to a user input to merge the one or more text segments into a single text segment for display; and if the user determines by viewing a display containing the at least one single text segment that the at least one single text segment should be split, perform a split action in response to the user input to split the at least one single text segment into the one or more text segments.