Caption Alignment via Anchor Word Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual captioning methods for live television or radio shows introduce delays, causing synchronization issues and inefficiencies, especially in real-time environments, and offline captioning is time-consuming and expensive.
Innovation Solution
A system utilizing automatic speech recognition (ASR) output and human-generated transcription to select anchor words for aligning captions, filtering out common words, and updating ASR dictionaries, which generates accurate and synchronized captions by delaying media playback to align with transcription.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual transcription is performed in real time or near real time, then captioning can be provided for live broadcasts, but delays of several seconds or longer occur causing synchronization issues
Solution Approach 1:
The system performs preliminary alignment between ASR output and transcription by identifying anchor words and establishing time mappings before final caption generation. This preliminary action allows the system to pre-calculate synchronization points, reducing real-time delays while maintaining accuracy.
Solution Approach 2:
The patent introduces an intermediary alignment process that uses anchor words as mediators between ASR output and transcription. These anchor words serve as reference points that bridge the timing gap between automated speech recognition and human transcription, enabling synchronized captioning without significant delays.
2Measurement precision
If human transcription is performed offline for pre-produced media, then instantaneous alignment is not necessary and accuracy can be improved, but the process becomes very time-consuming and expensive
Solution Approach 1:
The system uses automated speech recognition output to identify potential anchor words and perform initial alignment, reducing the manual work required for transcription. The ASR output serves itself by providing a framework that human transcribers can build upon, improving accuracy while reducing time and cost.
Solution Approach 2:
Instead of requiring complete manual transcription for entire media files, the system performs partial manual verification only at anchor word positions. This partial action approach maintains high accuracy by having humans verify critical synchronization points while automated systems handle the bulk of the transcription work.
3Measurement precision
If anchor words are selected based on similarity threshold between ASR output and transcription, then alignment accuracy is improved, but common words are incorrectly selected as anchors
Solution Approach 1:
The system applies different selection criteria to different word positions and contexts. Common words are excluded from anchor word selection through filtering, while unique and distinctive words are prioritized. This local quality approach ensures that anchor words have high discriminative power for accurate alignment.
Solution Approach 2:
The anchor word selection process is dynamic, adjusting the similarity threshold and filtering criteria based on the specific characteristics of the media content. The system adapts to different languages, domains, and content types, optimizing anchor word selection for each specific case rather than using fixed criteria.
Data Source
AI summary
Disclosed herein are systems, computer-implemented methods, and tangible computer-readable storage media for captioning a media presentation. The method includes receiving automatic speech recognition (ASR) output from a media presentation and a transcription of the media presentation. The method includes selecting via a processor a pair of anchor words in the media presentation based on the ASR output and transcription and generating captions by aligning the transcription with the ASR output between the selected pair of anchor words. The transcription can be human-generated. Selecting pairs of anchor words can be based on a similarity threshold between the ASR output and the transcription. In one variation, commonly used words on a stop list are ineligible as anchor words. The method includes outputting the media presentation with the generated captions. The presentation can be a recording of a live event.