Phase Calculator Using Target Envelope for Audio Signal Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio signal processing techniques suffer from audible artifacts such as musical noise, phase interference, and pre-echos during music source separation, particularly in transient signal reconstruction and score-informed audio decomposition.
Innovation Solution
The use of a target time-domain envelope to correct phase information in audio signals through iterative algorithms, where the phase is restored by maintaining or updating the amplitude, thereby reducing the number of iterations required and improving the quality of phase restoration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If iterative phase restoration algorithms are used to reconstruct time-domain signals from frequency-domain frames, then phase information can be recovered, but the process requires many iterations which increases computational complexity and time
Solution Approach 1:
The patent applies a target time-domain envelope to the frequency-domain frames before initiating the iterative phase restoration process. This preliminary action provides the algorithm with prior information about the expected signal characteristics, enabling faster convergence and reducing the number of iterations needed to achieve accurate phase restoration.
Solution Approach 2:
The target time-domain envelope serves as an intermediary that bridges the frequency-domain representation and the time-domain reconstruction. By incorporating this envelope information into the phase restoration process, the algorithm can more efficiently converge to the correct phase values without requiring excessive iterations.
2Reliability
If standard inverse transform methods are used to reconstruct time-domain signals from modified frequency-domain representations, then signal reconstruction is achieved, but audible artifacts such as pre-echos and smeared transients occur
Solution Approach 1:
The patent applies the target time-domain envelope to the magnitude spectrum before performing the inverse transform. This preliminary modification ensures that the reconstructed signal's envelope matches the target, thereby preventing artifacts like pre-echos and transient smearing from occurring in the first place.
Solution Approach 2:
The patent modifies the frequency-domain representation by applying the target envelope to the magnitude spectrum, changing the spectral parameters before reconstruction. This parameter modification ensures that the time-domain signal reconstructed from the modified frequency-domain data will have the desired envelope characteristics and avoid artifacts.
3Measurement precision
If score-informed decomposition techniques are used to separate music sources, then separation accuracy is improved, but the complexity of the processing increases
Solution Approach 1:
The patent incorporates score information (temporal and spectral constraints) as a preliminary guide for the source separation process. By using this prior knowledge to constrain the decomposition, the system achieves accurate separation without requiring overly complex iterative optimization procedures.
Solution Approach 2:
The target time-domain envelope acts as an intermediary that integrates score information with the frequency-domain decomposition. This envelope provides a simplified representation that guides the separation process while maintaining computational efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Subject of the invention is an apparatus (2) described by a schematic block diagram for processing an audio signal (4) to obtain a processed audio signal (6). The apparatus (2) comprises a phase calculator (8) for calculating phase values (10) for spectral values of a sequence of frequency-domain frames (12) representing overlapping frames of the audio signal (4). Moreover, the phase calculator 8 is configured to calculate the phase values (10) based on information on a target time-domain envelope (14) related to the processed audio signal (6), so that the processed audio signal (6) has at least in an approximation the target time-domain envelope (14) and a spectral envelope determined by the sequence of frequency-domain frames (12).