Voice Recognition Intent Probability Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice recognition systems face challenges in accurately interpreting user intents from voice signals, particularly in dynamic user and device environments, leading to inefficiencies in feature extraction and task execution.
Innovation Solution
A voice recognition apparatus that calculates a probability distribution for user intents using pre-stored training data, selects either a training-based or rule-based feature extractor based on the distribution, and adjusts probability values using user and device information to optimize feature extraction and task performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single feature extractor is used for all intents, then the device complexity is reduced, but the accuracy of intent recognition deteriorates
Solution Approach 1:
The patent segments the feature extraction process by dividing intents into different categories (e.g., information extraction, command execution) and assigning different feature extractors to each category. This allows each extractor to be optimized for its specific intent type, improving overall accuracy without requiring a single complex universal extractor.
Solution Approach 2:
The patent implements dynamic selection of feature extractors based on the detected intent type. The system transitions from a static single-extractor architecture to a dynamic multi-extractor system where the appropriate extractor is chosen at runtime, balancing complexity management with improved recognition accuracy.
2Measurement precision
If multiple feature extractors are used for different intents, then the intent recognition accuracy is improved, but the device complexity increases
Solution Approach 1:
The patent applies local quality by tailoring the characteristics of each feature extractor to match the specific requirements of its target intent type. Each extractor is designed with localized optimization for its specific domain, allowing high accuracy for each intent category while maintaining manageable complexity through specialization.
Solution Approach 2:
The patent creates a universal intent recognition system that can handle multiple intent types through a unified architecture that selects from multiple specialized extractors. The commonality lies in the selection mechanism and the overall framework, while the specialization is in the individual extractors.
3Measurement precision
If training-based feature extraction is used, then the recognition accuracy for complex intents is improved, but the computational time increases
Solution Approach 1:
The patent applies partial action by using training-based extractors only for complex intents that require deep learning, while using simpler rule-based extractors for straightforward intents. This selective application of computational resources ensures high accuracy where needed while minimizing unnecessary computational time for simpler tasks.
Solution Approach 2:
The patent changes the extraction method parameter based on intent complexity and type. The system dynamically adjusts whether to use training-based or rule-based extraction, effectively changing the computational approach to match the task requirements and optimize the balance between accuracy and speed.
4Loss of time
If rule-based feature extraction is used, then the computational time is reduced, but the accuracy for ambiguous intents deteriorates
Solution Approach 1:
The patent applies partial action by using rule-based extractors for clear, unambiguous intents where rules suffice, avoiding the computational overhead of training-based methods. For ambiguous or complex intents, the system switches to training-based extractors, ensuring accuracy is maintained where rules are insufficient.
Solution Approach 2:
The patent changes the extraction method parameter based on intent clarity and complexity. The system dynamically selects between rule-based and training-based approaches, adjusting the computational strategy to match the difficulty of the input intent and optimize both speed and accuracy.
Data Source
AI summary
A voice recognition apparatus and corresponding method include a processor configured to calculate a probability distribution corresponding to an intent associated with an utterance of a user by applying pre-stored training data to an input voice signal input based on the utterance. The processor is also configured to select a target feature extractor including either one or both of a training-based feature extractor and a rule-based feature extractor using the calculated probability distribution, and extract a feature associated with the utterance based on the selected target feature extractor.