Voice Recognition Domain Detection via Hierarchical Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional voice recognition systems often misinterpret user intentions due to their inability to consider multiple domains in a user's utterance, leading to unintended response information being provided, which requires further detailed input from the user to correct.
Innovation Solution
A dialogue-type voice recognition apparatus that extracts user action and object elements from an utterance voice, uses a hierarchical domain model with multi and binary classifiers to determine the intended domain, and transmits relevant information to an external apparatus for accurate response generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a conventional voice recognition apparatus detects only one domain from among multiple domains without considering multiple domains, then the device complexity is reduced, but the reliability of domain determination deteriorates
Solution Approach 1:
The system dynamically adjusts the domain detection process by first performing a broad detection across multiple domains, then selectively expanding only into candidate domains that show relevance to the user's utterance. This dynamic approach allows the system to handle multiple domains without permanently increasing structural complexity.
Solution Approach 2:
The domain detection process is segmented into distinct stages: initial domain detection, candidate domain identification, and selective expansion. By dividing the complex task of multi-domain detection into manageable segments, the system achieves high reliability without overwhelming device complexity.
2Productivity
If a voice recognition apparatus provides response information based on an arbitrarily determined domain, then the productivity of response generation is improved, but the reliability of user intent understanding deteriorates
Solution Approach 1:
The system performs preliminary domain analysis by identifying candidate domains before generating responses. This preliminary action ensures that the selected domain truly reflects user intent, preventing arbitrary domain selection while maintaining efficient response generation through pre-identified candidates.
Solution Approach 2:
The system incorporates feedback mechanisms where the detected domain and candidate domains inform the response generation process. This feedback loop ensures that responses are generated based on accurate domain understanding rather than arbitrary selection, improving both reliability and productivity.
3Reliability
If a voice recognition apparatus requires further detailed utterance input from the user to correct unintended responses, then the reliability of domain determination is improved, but the loss of time increases
Solution Approach 1:
The system performs preliminary domain detection and candidate identification before generating responses, ensuring that the first response is likely to be correct. This preliminary action reduces the need for corrective user input, minimizing time loss while maintaining high domain determination accuracy.
Data Source
AI summary
A voice recognition apparatus and a method for providing response information are provided. The voice recognition apparatus according to the present disclosure includes an extractor configured to extract a first utterance element representing a user action and a second utterance element representing an object from a user's utterance voice signal; a domain determiner configured to detect an expansion domain related to the extracted first and second utterance elements based on a hierarchical domain model, and determine at least one candidate domain related to the detected expansion domain as a final domain; a communicator which performs communication with an external apparatus; and a controller configured to control the communicator to transmit information regarding the first and second utterance elements and information regarding the determined final domain.