Domain-Specific Speech Recognition for Enterprise Meeting Action Items

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Manual processing of meeting content, such as action items and minutes, is time-consuming and requires significant human effort, especially in enterprise settings where different styles and concepts are involved, making it inefficient to capture, track, and manage meeting information.

Innovation Solution

Implementing a computer-implemented method that uses speech recognition with domain-specific training data and a ranking algorithm to automatically identify and display action items from audio, enhancing precision and relevance through a knowledge base and user feedback, allowing for real-time processing and management of meeting content.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If manual processing is used to capture and track meeting content, then flexibility in handling different styles and concepts is maintained, but time consumption and human effort increase significantly

Engineering Contradiction:

Improveflexibility in handling different styles and conceptsVSAvoidtime consumption

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent replaces manual mechanical processing of meeting content with an automated speech recognition system that uses neural networks and domain-specific knowledge bases to transcribe and analyze audio, eliminating the need for manual note-taking while maintaining adaptability through context-aware processing

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system adapts to different meeting styles and concepts by dynamically adjusting recognition parameters based on domain-specific knowledge bases and contextual information, allowing the same automated system to handle diverse content types without requiring manual reconfiguration

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If manual processing is used to organize and manage meeting minutes, then accuracy in capturing specific terminology is maintained, but productivity decreases due to repetitive tasks

Engineering Contradiction:

Improveaccuracy in capturing terminologyVSAvoidproductivity

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The speech recognition system performs self-calibration by automatically learning domain-specific terminology and meeting styles from training data and feedback, eliminating the need for manual processing while maintaining high accuracy through continuous improvement of its own performance

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system incorporates feedback mechanisms where users can correct recognized terms, and these corrections are fed back into the system to improve future recognitions, ensuring continuous improvement of accuracy without requiring manual processing for each individual task

Inventive Principle:

Principle #23Feedback

3Productivity

If automated speech recognition is used to process meeting audio, then productivity increases by reducing manual effort, but precision decreases due to lack of domain-specific context

Engineering Contradiction:

ImproveproductivityVSAvoidprecision of speech recognition

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary training using domain-specific knowledge bases and meeting transcripts before actual use, pre-loading contextual information about industry terminology, meeting structures, and common phrases to enable high-precision recognition from the start

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces domain-specific knowledge bases as intermediary components between the general speech recognition system and the meeting content, providing contextual bridges that enable accurate recognition of specialized terminology while maintaining automated processing

Inventive Principle:

Principle #24Intermediary (Mediator)

4Measurement precision

If domain-specific training data is provided to enhance speech recognition, then precision and relevance of action item identification improve, but device complexity increases

Engineering Contradiction:

Improveprecision of action item identificationVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The domain-specific knowledge bases are designed as universal, reusable components that can be applied across multiple meetings and domains, allowing the same trained model to handle various meeting types without requiring separate complex processing systems for each domain

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS9336776B2Enhancing speech recognition with domain-specific knowledge to detect topic-related content

Publication Date: 2016.05.10 SAP SE

AI summary

Methods, systems, and computer-readable storage media for providing action items from audio within an enterprise context. In some implementations, actions include determining a context of audio that is to be processed, providing training data to a speech recognition component, the training data being provided based on the context, receiving text from the speech recognition component, processing the text to identify one or more action items by identifying one or more concepts within the text and matching the one or more concepts to respective transitions in an automaton, and providing the one or more action items for display to one or more users.