Sound Analysis Apparatus for Spatial Noise Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice recognition systems face errors due to variations in noise characteristics and types across different spatial sound environments, as they often rely on a single noise model that fails to account for these changes.
Innovation Solution
A sound analysis method and apparatus that extracts and learns repeated sound patterns from a target space by dividing input sounds into sub-sounds, determining matching relationships, and training a sound learning model specific to that environment, using a processor and microphone to improve voice recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single noise model is used for voice recognition, then the system complexity is reduced, but the voice recognition accuracy deteriorates in varying sound environments
Solution Approach 1:
The patent segments the single noise model into multiple noise models corresponding to different spatial sound environments. Each noise model is trained on sound data collected from specific locations, allowing the system to select and apply the appropriate model based on the current environment, thereby maintaining accuracy without requiring a single overly complex universal model.
Solution Approach 2:
The patent implements a dynamic noise model selection mechanism that adapts to changing spatial environments. The system determines the current spatial sound environment and dynamically selects the corresponding noise model, enabling the voice recognition system to adjust to varying acoustic conditions rather than using a static single model.
2Reliability
If multiple noise models are created for different spatial environments, then the voice recognition accuracy is improved, but the device complexity increases
Solution Approach 1:
The patent implements an automatic noise model selection mechanism where the system autonomously determines the current spatial sound environment and selects the appropriate pre-trained noise model without requiring manual intervention. This self-service approach manages the complexity of multiple models by automating the selection process based on environmental characteristics.
Solution Approach 2:
The patent uses spatial environment parameters (such as location identifiers or acoustic characteristics) to select among different noise models. By changing the selection parameter based on the spatial environment, the system efficiently manages multiple noise models without requiring complex decision-making logic, simply matching environmental parameters to corresponding models.
3Adaptability or versatility
If noise models are trained for specific spatial environments, then the adaptability to sound environments is improved, but the training time and data collection requirements increase
Solution Approach 1:
The patent performs preliminary training of multiple noise models in advance for different spatial sound environments before actual voice recognition operations. By pre-training the models and storing them, the system avoids the need for real-time training when deployment is needed, thus reducing operational training time while maintaining high adaptability to various sound environments.
Data Source
AI summary
Disclosed is a sound analysis method and apparatus which execute an installed artificial intelligence (AI) algorithm and/or a machine learning algorithm and are capable of communicating with other electronic devices and servers in a 5G communication environment. The sound analysis method and apparatus provide a sound learning model specialized for a sound environment of a target space.