Voice Interaction Topic Shift via Prosodic Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice interaction robots struggle to change topics at appropriate timings due to reliance on syntactic analysis, which is time-consuming and requires specific user prompts, leading to potential delays or inappropriate topic changes.
Innovation Solution
A voice interaction apparatus and method that utilize non-linguistic information analysis, including prosodic and response history data, to determine topic continuation or change, allowing for timely topic shifts without syntactic analysis and user prompts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If syntactic analysis is used to determine topic continuation, then topic change accuracy is improved, but response time increases and topic changes are delayed
Solution Approach 1:
The patent segments the analysis of user speech into two independent parts: linguistic information analysis (syntactic analysis) and non-linguistic information analysis (prosodic analysis). These two analyses are performed in parallel rather than sequentially, allowing the system to determine topic continuation based on both linguistic and non-linguistic features simultaneously, thereby reducing overall processing time while maintaining accuracy.
Solution Approach 2:
The patent introduces prosodic information as an intermediary indicator that can signal topic change intentions without requiring complete syntactic analysis. Prosodic features such as pitch contours, pause patterns, and speech rate serve as intermediate cues that help the system detect topic transitions faster, acting as a mediator between raw speech input and final topic determination.
2Measurement precision
If syntactic analysis with user prompts is used, then topic change precision is improved, but user interaction complexity increases
Solution Approach 1:
The patent enables the system to automatically detect topic changes through prosodic analysis without requiring users to provide specific prompts or follow complex interaction patterns. The system self-services by analyzing inherent prosodic features in natural speech such as pitch variations and pause patterns, eliminating the need for users to learn or remember specific trigger words or phrases for topic changes.
Solution Approach 2:
The patent changes the detection parameters from linguistic features (requiring specific prompt words) to non-linguistic prosodic features (pitch, pause, speech rate). This parameter change allows the system to detect topic transitions based on natural variations in speech delivery rather than requiring specific lexical content, thereby simplifying user interaction while maintaining detection precision.
3Measurement precision
If only linguistic information analysis is used, then topic understanding accuracy is improved, but response speed decreases
Solution Approach 1:
The patent segments the speech analysis process into parallel linguistic and non-linguistic information processing streams. By dividing the analysis into independent segments that can be processed simultaneously, the system maintains comprehensive topic understanding through linguistic analysis while accelerating overall response speed through parallel prosodic analysis.
Solution Approach 2:
The patent applies partial action by using prosodic information as a supplementary indicator that can independently signal topic changes without requiring complete linguistic analysis. In cases where prosodic features strongly indicate a topic transition, the system can respond faster by relying partially on these non-linguistic cues rather than waiting for full syntactic parsing to complete.
Data Source
AI summary
A syntactic analysis unit 104 performs a syntactic analysis for linguistic information on acquired user' speech (hereinafter simply referred to as “user speech”). A non-linguistic information analysis unit 106 analyzes non-linguistic information different from the linguistic information for the acquired user speech. A topic continuation determination unit 110 determines whether a topic of the current conversation should be continued or should be changed to a different topic according to the non-linguistic information analysis result. A response generation unit 120 generates a response according to a result of a determination by the topic continuation determination unit 110.