Conversation Group Detection for Hearing Assistance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hearing-impaired individuals face difficulties in following conversations in noisy environments due to existing systems requiring manual selection of interlocutors, which becomes tedious when interlocutors change or participate in multiple conversations, leading to distraction and reduced understanding.
Innovation Solution
A method using computer equipment connected to voice transmission modules, display devices, and voice activity detection systems to automatically determine and manage conversation groups, allowing for seamless tracking of relevant conversations without manual intervention, utilizing voice activity, face orientation, and gaze detection to correlate interlocutor participation and adjust audio or text output accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual selection of interlocutors is used to filter conversations, then the hearing-impaired person can focus on specific conversations, but the system requires continuous manual correction when interlocutors change or join/leave conversations
Solution Approach 1:
The system automatically detects and tracks conversation groups by analyzing voice activity patterns and temporal correlations between interlocutors, eliminating the need for manual selection. The algorithm self-adjusts when interlocutors join or leave conversations by continuously monitoring speech patterns and correlation metrics.
Solution Approach 2:
The system uses feedback from voice activity detection and temporal correlation analysis to automatically update conversation group memberships. When speech patterns indicate a change in conversation participants, the system adjusts the grouping without user intervention, maintaining reliable conversation following.
2Reliability
If the hearing-impaired person manually corrects interlocutor selection regularly, then conversation following accuracy can be maintained, but the attention and effort required prevent full participation in the conversation
Solution Approach 1:
The system performs preliminary automatic grouping of interlocutors based on voice activity detection before the user needs to follow the conversation. Conversation groups are established and maintained in advance through continuous analysis of speech patterns, so the user can immediately begin following without manual setup or correction.
Solution Approach 2:
The conversation tracking system continuously self-updates by detecting changes in voice activity patterns and temporal correlations, automatically maintaining accurate conversation groups without requiring the user to spend time on manual corrections or adjustments.
3Loss of information
If all interlocutor voices are transmitted to the hearing-impaired person, then complete conversation information is available, but extraneous noises from multiple conversations make it difficult to follow a single conversation
Solution Approach 1:
The system extracts and isolates the voices of interlocutors belonging to the selected conversation group from the mixed audio environment. By identifying temporal correlations and voice activity patterns specific to the target conversation, the system separates desired speech signals from extraneous noises of other conversations.
Solution Approach 2:
The audio environment is segmented into distinct conversation groups based on temporal correlation analysis of interlocutor speech patterns. The system divides the continuous audio stream into separate conversation channels, allowing the user to select and follow one conversation while filtering out others.
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a method for assisting a hearing-impaired person equipped with a computer device in following a conversation involving a plurality of speakers, comprising the steps of: acquiring (E1, E2) signals representative of the voice activity of a first speaker and characterising the behaviour of a second speaker in response to the voice activity of the first speaker; determining (E3) if the first and second speakers belong to the same first conversation group on the basis of the signals acquired; selecting (E4) the first conversation group; for the first conversation group, determining (E5) a voice playback mode for the emission (E6) of the voice signals acquired for the speakers belonging to the first conversation group or a text delivery mode for the display (E7) of text signals obtained by converting the voice signals acquired for the speakers belonging to the first conversation group.