Audio Categorization Using Gaussian Mixture Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Electronic devices struggle to accurately identify and categorize audio signals in environments with multiple sound sources, leading to improper execution of predetermined functions.
Innovation Solution
An audio categorization apparatus and method that utilizes a storage circuit and categorization processing circuit to perform audio framing, feature extraction, and Gaussian mixture model training on training audio files, generating Gaussian curves and curve parameters for categorizing features, which are then used to identify and categorize input audio signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If electronic apparatuses receive various kinds of sounds from multiple sound sources, then the audio input becomes more complex and diverse, but the ability to accurately identify and categorize audio signals deteriorates
Solution Approach 1:
The patent segments audio signals into distinct categories by extracting specific acoustic features (such as spectral characteristics, temporal patterns, and frequency distributions) and comparing them against pre-defined category profiles. This segmentation approach allows the system to differentiate between various sound sources and categories even in complex acoustic environments with multiple simultaneous sound sources.
Solution Approach 2:
The patent utilizes parameter changes in the acoustic signal domain by analyzing variations in frequency, amplitude, temporal patterns, and spectral characteristics to identify and categorize different audio signals. By monitoring changes in these acoustic parameters over time, the system can distinguish between different sound sources and categories, maintaining high identification accuracy despite the presence of multiple simultaneous sound sources.
2Reliability
If audio categorization mechanism is not equipped, then the device structure remains simple, but the execution timing of predetermined functions becomes improper
Solution Approach 1:
The patent implements preliminary action by pre-defining audio categories and their corresponding characteristic profiles before actual audio categorization is needed. The system pre-processes and stores category-specific acoustic feature patterns, allowing for rapid and accurate real-time categorization without requiring complex processing during function execution. This preliminary preparation ensures reliable function execution timing while keeping the real-time processing mechanism relatively simple.
Data Source
AI summary
An audio categorization method is provided that includes steps outlined below. From a plurality of training audio files categorized into a plurality of audio categories, one of the audio categories is selected to be a corresponding audio category and the training audio files categorized in to the corresponding audio category is retrieved so as to perform audio framing and feature extraction thereon to generate a plurality of training feature data. A Gaussian mixture model training is performed on the training feature data to generate a plurality of Gaussian distribution curves to approximate a data distribution of the training feature data. A plurality of curve parameters of the Gaussian distribution curves are generated to be a categorizing feature of the corresponding audio category.