Ambience Extraction Using Correlation-Based Time-Frequency Masks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting ambience from audio signals suffer from suboptimal performance due to leakage of dominant sources and underestimation of ambience caused by short-term estimation of cross-correlation coefficients, particularly when dominant sources are panned to either channel.
Innovation Solution
The method employs a time-frequency analysis-synthesis framework to derive ambience extraction masks based on signal correlations, compensating for biases in short-term cross-correlation estimates and assuming equal ratios or levels of ambience in channels to improve ambience extraction, using a system with modules for time-frequency transformation, correlation computation, mask derivation, and multiplication.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If short-term estimation of cross-correlation coefficient is used, then the method can operate in real-time, but the amount of ambience is underestimated
Solution Approach 1:
The patent applies parameter changes by modifying the cross-correlation coefficient estimation process. Specifically, it transforms the estimated cross-correlation coefficient through a mathematical function (taking the square root of the sum of squares of real and imaginary parts) to compensate for the underestimation bias inherent in short-term estimation, thereby improving measurement precision while maintaining real-time operation capability
Solution Approach 2:
The patent replaces the direct use of short-term cross-correlation estimates with a compensated version derived from time-frequency analysis. By substituting the biased mechanical estimation process with a corrected mathematical transformation based on spectral correlations, the system achieves more accurate ambience quantification without sacrificing computational efficiency
2Ease of manufacture
If ad hoc functions are used for determining ambience extraction masks, then the implementation is simple, but the extraction performance is suboptimal with significant leakage of dominant sources
Solution Approach 1:
The patent changes the parameters used for mask determination from ad hoc functions to correlation-based parameters. By deriving masks from the magnitude of compensated cross-correlation coefficients and comparing them against thresholds, the system achieves more accurate separation of ambience from dominant sources while maintaining implementation feasibility through established signal processing operations
Solution Approach 2:
The patent introduces correlation quantities as an intermediary between the input signals and the ambience extraction masks. This intermediary step provides a more reliable basis for mask determination by capturing the statistical relationship between channels, thereby reducing leakage of dominant sources while keeping the overall system structure manageable
3Measurement precision
If time-frequency analysis-synthesis framework is used, then primary and ambient components are resolved accurately, but the computational complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the audio signal into time-frequency segments using short-time Fourier transform. This allows independent analysis of correlations in different time and frequency regions, improving component separation accuracy while managing computational complexity through localized processing rather than global analysis
Solution Approach 2:
The patent implements local quality by applying different processing strategies to different time-frequency regions. By computing correlations and deriving masks locally in each frequency band and time frame, the system achieves high precision in component separation while avoiding the computational burden of uniform high-resolution processing across the entire signal
Data Source
AI summary
A method of ambience extraction includes analyzing an input signal to determine the time-dependent and frequency-dependent amount of ambience in the input signal, wherein the amount of ambience is determined based on a signal model and correlation quantities computed from the input signals and wherein the ambience is extracted using a multiplicative time-frequency mask. Another method of ambience extraction includes compensating a bias in the estimation of a short-term cross-correlation coefficient. In addition, systems having various modules for implementing the above methods are disclosed.