Voice Dialogue System Utterance Intent Determination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice dialogue systems struggle to accurately determine whether a user utterance overlapping with a system utterance is a response to the current system utterance.
Innovation Solution
Incorporating a tag into the system utterance, allowing the intention understanding unit to determine the user's utterance intention based on whether the user's utterance timing is before or after the output of the corresponding system utterance sentence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system uses existing methods to determine user utterance intentions, then the system can identify directed user utterances, but it cannot accurately determine whether overlapping user utterances are responses to current system utterances
Solution Approach 1:
The system utterance is segmented into a connective portion and a content portion, with the tag positioned between them. This segmentation allows the intention understanding unit to distinguish between user utterances that overlap with the connective portion (not responses to current system utterance) and those that overlap with the content portion (responses to current system utterance), thereby improving determination accuracy without requiring complex analysis of the entire utterance structure
Solution Approach 2:
A tag is introduced as an intermediary element between the connective portion and content portion of the system utterance. This tag serves as a mediator that provides temporal reference information, enabling the intention understanding unit to determine whether a user utterance is a response to the current system utterance by comparing the user utterance timing with the tag position, thus improving measurement precision with minimal added complexity
2Ease of operation
If the system outputs complete system utterances before determining user response, then the dialogue flows naturally, but the system cannot respond appropriately to overlapping user utterances
Solution Approach 1:
The tag is embedded in the system utterance text before the utterance is output by the voice output unit. This preliminary action allows the intention understanding unit to have advance knowledge of the temporal reference point, enabling it to accurately determine whether overlapping user utterances are responses to the current system utterance, thus improving response appropriateness while maintaining natural dialogue flow
Data Source
AI summary
A voice dialogue system includes: a voice input unit which acquires a user utterance, a dialogue text creator which creates a text of a system utterance, a voice output unit which outputs the system utterance as voice data, and a setting unit that sets a response deadline for system utterances. When a first system utterance is output by the voice output unit, a second system utterance is output after the first system utterance without having acquired a user utterance, and the second system utterance comprises a connective portion for connecting following sentences and a content portion that is a subject of the second system utterance, the setting unit sets a timing at which output of the connective portion of the second system utterance ends or output of the content portion of the second system utterance starts as the response deadline for the first system utterance.