Content-Aware Voice Recognition Without a Call Word
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic devices with voice recognition functions require a call word for operation, leading to malfunctions when recognizing user utterances unrelated to device functions.
Innovation Solution
An electronic apparatus and method that identifies a dedicated language model based on displayed content, recognizes user utterances, and evaluates similarity and suitability using a geometric mean of entropy to determine relevant operations without a call word, employing a filtering language model to ignore irrelevant utterances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the electronic apparatus operates by recognizing user utterance without a call word, then the ease of operation is improved, but malfunctions occur due to performing operations for utterances unrelated to device functions
Solution Approach 1:
The patent changes the parameter of language model selection by switching from a general-purpose language model to a dedicated language model tailored to specific displayed content. This allows the system to maintain high ease of operation (no call word needed) while improving reliability by filtering out irrelevant utterances through content-specific language understanding.
Solution Approach 2:
The patent introduces a dedicated language model as an intermediary between the voice recognition system and the control system. This intermediary layer processes and evaluates candidate texts against the displayed content context, enabling the system to distinguish between relevant and irrelevant utterances without requiring a call word, thus resolving the contradiction between ease of operation and reliability.
2Adaptability or versatility
If a general-purpose language model is used for voice recognition, then the adaptability is improved, but the precision of identifying relevant utterances deteriorates
Solution Approach 1:
The patent changes the parameter of language model specificity by selecting a dedicated language model that is tailored to the displayed content rather than using a general-purpose model. This improves measurement precision in identifying relevant utterances while maintaining adaptability through the ability to switch models based on content type.
Solution Approach 2:
The patent applies local quality by using different dedicated language models for different types of displayed content (e.g., video playback vs. image display). Each content type has its own optimized language model, improving precision for that specific context while the overall system remains adaptable to various content types.
3Reliability
If the electronic apparatus requires a call word for voice recognition, then the reliability is improved by preventing malfunctions, but the ease of operation deteriorates
Solution Approach 1:
The patent changes the parameter of language model dedication from general-purpose to content-specific, enabling the system to achieve high reliability without requiring a call word. The dedicated language model provides contextual understanding that filters irrelevant utterances, maintaining reliability while improving ease of operation.
Solution Approach 2:
The dedicated language model serves as an intermediary that enables call-word-free operation while maintaining reliability. It evaluates candidate texts against the displayed content context, allowing the system to reliably distinguish relevant from irrelevant utterances without the need for a call word trigger.
4Measurement precision
If a dedicated language model is used for specific displayed content, then the precision of identifying relevant utterances is improved, but the device complexity increases
Solution Approach 1:
The patent changes the parameter of language model selection to use dedicated models for specific content types, improving precision in identifying relevant utterances. The complexity is managed through systematic model selection based on content type rather than implementing complex real-time model generation or switching mechanisms.
Data Source
AI summary
Provided are an electronic apparatus and method for performing an operation based on recognizing a user's command utterance without a call word. The method includes identifying a dedicated language model related to a displayed content; receiving an utterance of a user; recognizing the received utterance and identifying candidate texts of the recognized utterance; identifying a similarity between the recognized utterance and the identified candidate texts; identifying, based on the identified dedicated language model and a predetermined threshold value, a suitability of a predetermined number of candidate texts with a high identified similarity, among the candidate texts; based on the identified suitability being outside a predetermined suitability range, ignoring the recognized utterance; and based on the identified suitability being in the predetermined suitability range, identifying a candidate text having a highest suitability, among the candidate texts, as the recognized utterance, and performing a corresponding operation.