Meeting Transcript Summarization with Participant Attribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for meeting summarization struggle with accurately summarizing lengthy and heterogeneous meeting transcripts, often introducing errors from automatic speech recognition and failing to identify and attribute content to specific participants, especially unknown ones, due to their inability to handle the unique semantic structures and styles of meeting data.
Innovation Solution
A computing system that processes electronic content from meetings, attributes it to contributing entities using user profiles, and generates improved summaries and templates by post-processing transcripts to correct errors and enhance readability, while identifying and attributing speech tags to specific participants, even unknown ones, through a combination of machine learning models and speech recognition technologies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing deep neural models for text summarization are used, then abstractive summaries can be generated, but they fail to handle the semantic structures and styles of meeting transcripts
Solution Approach 1:
The patent applies local quality by customizing the summarization model specifically for meeting transcripts rather than using generic document summarization models. The system adapts the model's semantic understanding to handle the unique structures of meeting data including multiple speakers, turns, and participant interactions, thereby improving both adaptability and accuracy for this specific domain.
Solution Approach 2:
The patent changes key parameters of the summarization model including training data composition (using meeting transcripts instead of general documents), model architecture adjustments for handling multi-speaker data, and optimization objectives tailored to meeting summarization tasks. These parameter changes enable the model to accurately capture meeting-specific semantic patterns.
2Measurement precision
If manual meeting minutes preparation is assigned to one participant, then accurate summaries can be created, but the process is time-consuming and not scalable
Solution Approach 1:
The patent replaces the mechanical manual process of meeting minutes preparation with an automated deep learning-based natural language processing system. The model automatically processes meeting transcripts, identifies key information, and generates summaries without human intervention, thereby maintaining high accuracy while dramatically improving efficiency and scalability.
Solution Approach 2:
The system enables self-service by allowing meeting transcripts to be automatically processed and summarized without requiring human participants to manually create minutes. The automated model independently handles the entire summarization workflow from input transcript to output summary.
3Productivity
If ASR systems are used to transcribe meetings, then large volumes of meeting transcripts can be generated, but transcription errors are introduced that propagate through downstream operations
Solution Approach 1:
The patent implements feedback mechanisms where the summarization model is trained to be robust against ASR errors and can identify and correct transcription mistakes during the summarization process. The system uses feedback from the context and semantic structure to detect and correct errors, thereby maintaining high reliability even when processing ASR-generated transcripts with errors.
4Adaptability or versatility
If multi-stage machine learning pipelines are used for abstractive summarization, then comprehensive processing can be achieved, but the pipelines are not end-to-end optimizable and require extensive human involvement
Solution Approach 1:
The patent merges multiple separate processing stages into a unified end-to-end deep learning model. Instead of using separate modules for different processing tasks that require manual tuning and optimization, the system combines these functions into a single trainable model that can be optimized globally, reducing complexity while maintaining comprehensive processing capabilities.
Solution Approach 2:
The patent creates a universal summarization model that handles multiple functions including error correction, semantic understanding, information extraction, and summary generation within a single framework. This multi-functional approach eliminates the need for separate specialized modules and reduces overall system complexity.
Data Source
AI summary
Attributes of electronic content from a meeting are identified and evaluated to determine whether sub-portions of the electronic content should or should not be attributed to a user profile. Upon determining that the sub-portion should be attributed to a user profile, attributes of the sub-portion of electronic content are compared to attributes of stored user profiles. A probability that the sub-portion corresponds to at least one stored user profile is calculated. Based on the calculated probability, the sub-portion is attributed to a stored user profile or a guest user profile.