Acoustic Time Series Conversion Accuracy via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating mathematical models that convert acoustic time series assume similarity between time series segments decreases with longer intervals, which is not always true, leading to low accuracy in conversion.
Innovation Solution
A learning device and method that includes a self-learning unit and a data augmentation unit, where the data augmentation unit performs clipping, duplication, and conversion processing on partial time series, and the self-learning unit updates the main conversion processing based on the results, allowing for improved accuracy in acoustic time series conversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If self-supervised learning is performed using pairs of segments from different times of one acoustic time series, then the learning process can be simplified, but the conversion accuracy of acoustic time series deteriorates when the assumption about similarity and time interval does not hold
Solution Approach 1:
The acoustic time series is divided into multiple segments, and pairs of segments are constructed by clipping from different times. This segmentation allows the learning algorithm to process data in manageable portions while maintaining the temporal relationships necessary for accurate conversion.
Solution Approach 2:
The patent introduces data augmentation processing that modifies parameters of the acoustic time series segments through clipping, duplication, and conversion operations. This creates varied training data that helps the model generalize better and improves conversion accuracy when the simple time-interval similarity assumption does not hold.
2Measurement precision
If data augmentation processing including clipping and duplication is performed on partial time series, then the conversion accuracy is improved, but the processing time and computational complexity increase
Solution Approach 1:
Data augmentation processing is performed in advance during the training phase, creating pre-transformed training data. This preliminary action allows the model to learn robust features during training, and during actual conversion tasks, the model can process data more efficiently without requiring real-time augmentation.
Solution Approach 2:
The patent uses duplication processing to create copies of partial time series segments and combines them through conversion operations. This copying strategy enables the model to learn from multiple versions of the same data pattern, improving generalization capability while managing computational load through efficient data reuse.
Data Source
AI summary
One aspect of the present invention is a learning device including a self-learning unit that updates content of main conversion processing for converting data to be processed into data in a predetermined format by executing self-supervised learning, and a data augmentation unit that executes data augmentation processing of generating data to be processed in the main conversion processing based on an acoustic time series, in which the data augmentation unit performs acoustic time series clipping processing of clipping a partial time series that is a time series of a part of the acoustic time series, duplication processing of duplicating the partial time series, and conversion processing of converting one and the other of the partial time series according to a predetermined rule, and the self-learning unit updates the content of the main conversion processing by self-supervised learning based on a result obtained by the conversion processing.