Monaural Vocal Separation Post-Processing for Instrument Harmonic Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional vocal separation techniques from monaural music recordings fail to effectively remove harmonics of pitched instruments from the separated vocal tracks due to their similar harmonic structures, leading to residual instrument sounds in vocal tracks.
Innovation Solution
A post-processing method using the Hough transform to detect and remove pitched instrument harmonics from the magnitude spectrum of separated vocals through adaptive median filtering, improving the quality of the vocal signal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional separation techniques are used to separate vocals from monaural music recordings, then the separation process is simple and fast, but the separated vocal tracks still contain harmonics of pitched instruments resulting in poor separation quality
Solution Approach 1:
The patent segments the separation process into two distinct stages: first applying traditional separation techniques to obtain initial vocal tracks, then applying a post-processing step using Hough transform to specifically detect and remove remaining instrument harmonics. This segmentation allows each stage to focus on specific aspects of the problem, improving overall separation quality without requiring complete redesign of the entire system.
Solution Approach 2:
The patent performs preliminary separation using traditional techniques before applying the more complex Hough transform-based refinement. By removing the bulk of separation work first, the subsequent post-processing step only needs to handle residual harmonics, making the overall process more efficient and manageable while achieving high separation quality.
2Manufacturing precision
If noise suppression techniques are used to separate vocals, then the processing is straightforward, but the techniques fail to separate pitched instruments since they do not possess whiteness and stationarity properties of noise
Solution Approach 1:
The patent introduces Hough transform as an intermediary detection mechanism that bridges the gap between traditional separation and effective harmonic removal. The Hough transform acts as a specialized detector that can identify the specific geometric patterns (horizontal ridges) of instrument harmonics in the spectrogram, enabling precise detection and removal of these components that traditional noise suppression cannot handle.
Solution Approach 2:
The patent changes the detection approach from time-domain noise suppression to frequency-time domain analysis using spectrograms and Hough transform. By transforming the problem into detecting horizontal lines in the spectrogram domain, the system can effectively identify and remove instrument harmonics based on their characteristic temporal and spectral patterns, overcoming the limitations of traditional noise suppression methods.
3Manufacturing precision
If post-processing techniques using Hough transform are applied to remove pitched instruments harmonics, then the separation quality improves significantly, but the processing time and computational complexity increase
Solution Approach 1:
The patent performs the computationally intensive Hough transform-based post-processing only on the residual harmonics that remain after traditional separation, rather than applying it to the entire signal from scratch. This preliminary separation approach reduces the amount of data and computational effort required for the refinement stage, minimizing additional processing time while maintaining high separation quality.
Solution Approach 2:
The patent applies the Hough transform-based harmonic removal selectively to specific regions in the spectrogram where instrument harmonics are detected (horizontal ridges), rather than uniformly processing the entire frequency spectrum. This localized approach focuses computational resources only where needed, improving separation quality in critical regions while reducing overall processing time compared to full-spectrum processing.
Data Source
AI summary
There is provided a post processing technique or method for separation algorithms to separate vocals from monaural music recordings. The method comprises detecting traces of pitched instruments in a magnitude spectrum of a separated voice using Hough transform and removing the detected traces of pitched instruments using median filtering to improve the quality of the separated voice and to form a new separated music signal. The method further comprises applying adaptive median filtering techniques to remove the identified Hough regions from the vocal spectrogram producing separated pitched instruments harmonics and new vocals while adding the separated pitched instruments harmonics to a music signal separated using any separation algorithm to form the new separated music signal.