Speech Processing Device Multi-Speaker Conversation Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques have low accuracy in extracting conversation groups of three or more speakers due to difficulties in detecting speech and establishing conversation between silent speakers and those who barely speak.
Innovation Solution
A speech processing device and method that detect speech from individual speakers, calculate degrees of established conversation in segments, and extract conversation groups based on long-time features, allowing for accurate identification of conversational partners in multi-speaker environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional technique based on sound-silent period alternation is used, then conversation groups of two speakers can be extracted, but accuracy drops significantly for conversation groups of three or more speakers
Solution Approach 1:
The determination time period is divided into multiple segments, and the degree of established conversation is calculated separately for each segment. This segmentation allows the system to detect conversation patterns across different time windows, improving accuracy in identifying conversation groups of three or more speakers by analyzing local conversation dynamics within each segment rather than relying solely on global sound-silent alternation patterns.
Solution Approach 2:
The patent introduces a new dimension of analysis by calculating the degree of established conversation not just based on sound-silent alternation, but by incorporating the temporal distribution of speech across multiple segments. This dimensional expansion from binary (sound/silent) to multi-dimensional (temporal distribution across segments) enables better detection of multi-speaker conversations where simple alternation patterns fail.
2Ease of manufacture
If degree of established conversation is calculated based on sound-silent alternation, then simple implementation is achieved, but low accuracy occurs when substantial listeners who barely speak are present
Solution Approach 1:
By segmenting the determination time period into multiple segments and calculating the degree of established conversation for each segment separately, the system can detect subtle conversation patterns that may be missed in a single global calculation. This allows substantial listeners who speak occasionally to be identified as conversational partners through their presence in specific segments, even though their overall speech duration is low.
Solution Approach 2:
The patent changes the parameter used for measuring established conversation from a simple sound-silent alternation metric to a segmented temporal distribution metric. This parameter change enables the system to capture the contribution of substantial listeners who speak intermittently, improving detection accuracy without requiring complex calculation methods.
Data Source
AI summary
A speech processing device which can accurately extract a conversation group from among a plurality of speakers, even when a conversation group formed of three or more people is present. This device (400) comprises: a spontaneous speech detection unit (420) and a direction-specific speech detection unit (430) which separately detect, from a sound signal, uttered speech from the speakers; a conversation establishment level calculation unit (450) which calculates a conversation establishment level for each separated segment of the time being determined, for all of the pairings of two people, on the basis of the detected uttered speech; an extended-period characteristic amount calculation unit (460) which calculates an extended-period characteristic amount for the conversation establishment level of the time being determined, for each pairing; and a conversation-partner determination unit (470) which extracts a conversation group which forms a conversation on the basis of the calculated extended-period characteristic amount.