Audio Separation Using Repeating Pattern Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing music/voice separation systems rely on specific audio features, complex frameworks, and require prior training, limiting their ability to accurately separate components from audio recordings without these features or with complex computational costs.
Innovation Solution
The Repeating Pattern Extraction Technique (REPET) algorithm identifies repeating structures in audio recordings by segmenting them based on temporal periods, generating a repeating segment model, and using a time-frequency mask to separate music and voice components without relying on specific features or complex frameworks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If known separation systems use specific audio features (MFCCs, pitch contour, chromagram), then separation accuracy is improved, but the system cannot handle recordings without these features and requires complex computational processing
Solution Approach 1:
The patent applies universality by creating a separation system that works with multiple types of audio features (spectrogram, MFCC, chromagram, pitch contour) rather than relying on a single feature type. The system can adapt to different feature representations and audio recording types, making it versatile while maintaining accuracy through feature-specific optimization.
Solution Approach 2:
The patent employs parameter changes by adjusting separation parameters based on the specific audio feature being used. Different features require different parameter settings for optimal separation, and the system dynamically adapts parameters to match the input feature type, thereby maintaining high accuracy across diverse audio recordings.
2Measurement precision
If known separation systems use complex frameworks, then separation performance is improved, but computational cost increases significantly
Solution Approach 1:
The patent applies segmentation by dividing the audio signal into smaller temporal segments and processing them independently. This reduces the computational burden of handling entire audio recordings at once while maintaining separation performance through localized analysis of each segment.
Solution Approach 2:
The patent uses partial action by applying separation only to specific time-frequency regions where separation is needed, rather than processing the entire spectrogram uniformly. This reduces computational cost by focusing processing resources on relevant portions of the audio signal.
3Measurement precision
If known separation systems require prior training, then separation accuracy is improved, but the system cannot process recordings without training data
Solution Approach 1:
The patent applies self-service by enabling the separation system to adapt to new audio types and features without requiring external training data. The system learns from the input audio itself and adjusts its parameters and models autonomously, allowing it to process unseen audio types while maintaining accuracy.
Data Source
AI summary
A method includes determining a first spectrogram of the audio signal, defining a similarity matrix of the audio signal based on the first spectrogram and a transposed version of the first spectrogram, identifying two or more similar frames in the similarity matrix that are more similar to a designated frame than to one or more other frames in the similarity matrix, creating a repeating spectrogram model based on the two or more similar frames that are identified in the similarity matrix, and deriving a mask based on the repeating spectrogram model and the first spectrogram of the audio signal. The mask is representative of similarities between the repeating spectrogram model and the first spectrogram of the audio signal. The method also includes extracting a repeating structure from the audio signal by applying the mask to the audio signal.