Emotion Detection Using Audio Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing emotion detection methods struggle with accurately determining emotional states in audio signals, especially when noise is present and fail to account for high-pitched voices or rapid emotional changes, leading to reduced accuracy and incorrect classifications.
Innovation Solution
An emotion detection method that uses both audio feature vectors and statistical models to analyze temporal transitions in emotional states, incorporating fundamental frequency, power, and speech rate variations, to improve accuracy and handle noisy conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a plurality of pieces of learning audio signal data are retained to increase determination accuracy, then the determination accuracy is improved, but the memory cost and calculation cost become enormous
Solution Approach 1:
The patent extracts only the essential features from audio signal data (fundamental frequency, power, temporal variation characteristics) and stores these compressed feature vectors rather than retaining the complete original audio data. This extraction process significantly reduces the amount of data that needs to be stored and processed, thereby reducing memory cost and calculation cost while maintaining determination accuracy.
Solution Approach 2:
The patent transforms the audio signal data into a different parameter space by extracting specific acoustic features (fundamental frequency, power, temporal variation) and representing them as feature vectors. This parameter transformation allows the system to work with a more compact and manageable representation of the audio data, reducing the complexity of storage and computation while preserving the essential information needed for accurate emotion determination.
2Productivity
If audio data is analyzed to extract fundamental frequency and power to determine emotional state, then the emotion detection is performed, but the determination accuracy is reduced when noise is present and high-pitched voices or rapid emotional changes are not accounted for
Solution Approach 1:
The patent incorporates temporal variation characteristics of the fundamental frequency and power into the feature extraction process. By analyzing how these acoustic parameters change over time rather than just their static values, the system can detect rapid emotional changes and adapt to dynamic speech patterns. This dynamic analysis enables the system to handle noisy conditions and detect high-pitched voices or rapid emotional transitions that static analysis would miss.
Solution Approach 2:
The patent adds the temporal dimension to the feature extraction by including temporal variation characteristics alongside the fundamental frequency and power. This dimensional expansion allows the system to capture dynamic information about how acoustic parameters evolve over time, providing additional cues for accurate emotion determination in noisy environments and for detecting rapid emotional changes or high-pitched voices.
Data Source
AI summary
An audio feature is extracted from audio signal data for each analysis frame and stored in a storage part. Then, the audio feature is read from the storage part, and an emotional state probability of the audio feature corresponding to an emotional state is calculated using one or more statistical models constructed based on previously input learning audio signal data. Then, based on the calculated emotional state probability, the emotional state of a section including the analysis frame is determined.