Audio Signal Segmentation Using LDA and Dynamic Programming
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing signal segmentation methods based on novelty functions often incorrectly identify segment boundaries due to peaking within segments as well as at boundaries, leading to inaccurate determinations.
Innovation Solution
The method segments audio signals by converting them into feature-space sequences using sliding window analysis, applying linear discriminant analysis (LDA) for conditioning, and employing dynamic programming to identify cluster boundaries, which are then used to determine segment boundaries, utilizing local, transition, and bias costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If novelty functions are used to estimate segment boundaries, then segment boundaries can be identified, but incorrect boundaries are identified due to peaking within segments as well as at boundaries
Solution Approach 1:
The patent segments the audio signal into characteristic regions by applying dynamic programming to divide the signal into segments with similar feature sets. This segmentation approach identifies boundaries between distinct audio regions by finding optimal segmentation points that minimize within-segment variation and maximize between-segment differences, thereby avoiding incorrect boundaries that novelty functions produce due to their sensitivity to transient peaks within segments.
Solution Approach 2:
The patent changes the parameter space by transforming audio signals into a feature space representation using spectral features and other audio characteristics. By operating in this transformed parameter space rather than the raw time-domain signal, the method can distinguish between meaningful segment boundaries and spurious peaks within segments, improving the reliability of boundary identification.
2Measurement precision
If dynamic programming is applied to feature vectors, then accurate segment boundaries are identified, but computational complexity increases
Solution Approach 1:
The patent performs preliminary action by first transforming the audio signal into a feature space representation and pre-processing the signal to extract relevant characteristics before applying dynamic programming. This preliminary feature extraction and signal conditioning reduces the dimensionality and complexity of the data that dynamic programming must process, making the computationally intensive algorithm more feasible while maintaining accurate segment boundary identification.
Data Source
AI summary
An input signal is converted to a feature-space representation. The feature-space representation is projected onto a discriminant subspace using a linear discriminant analysis transform to enhance the separation of feature clusters. Dynamic programming is used to find global changes to derive optimal cluster boundaries. The cluster boundaries are used to identify the segments of the audio signal.