Sound-Source Separation With Phase Difference Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In compact sound collector devices, the narrow disposing pitch of microphones results in small amplitude and phase differences between signals, making it difficult to achieve precise sound-source separation, especially in low-frequency ranges and high-frequency ranges with long wavelengths, and increasing the computational load for real-time directivity.
Innovation Solution
A sound-source separation method that involves filtering, interchanging, and updating coefficients to adjust input signals for equal amplitude and phase, using a specific time delay and recurrence formula to emphasize sound from a target direction with reduced computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Volume of moving object
If microphones are disposed closely in compact sound collector devices, then device size is reduced, but amplitude difference and phase difference between signals become very small making sound-source separation difficult
Solution Approach 1:
The patent applies parameter changes by using a phase difference calculator that computes phase differences in the frequency domain rather than directly in the time domain. This transformation allows for accurate detection of small phase differences even when microphones are closely disposed, resolving the contradiction between compact device size and measurement precision.
2Measurement precision
If sophisticated frequency analysis and complex analysis are applied to detect waveform structure differences, then sound-source separation accuracy is improved, but frame length and filter coefficients become very long increasing arithmetic processing load
Solution Approach 1:
The patent extracts only the essential phase difference information from the input signals using a phase difference calculator, rather than performing comprehensive sophisticated frequency analysis. This extraction approach maintains sound-source separation accuracy while significantly reducing the arithmetic processing load and avoiding the need for long frame lengths and complex filter coefficients.
Solution Approach 2:
The patent substitutes complex mechanical signal processing (long filters, extensive frequency analysis) with a simpler phase difference calculation mechanism. By replacing the complicated analysis system with a direct phase difference computation, the invention achieves accurate sound-source separation with much lower arithmetic processing requirements.
3Device complexity
If the number of microphones is increased to reduce arithmetic processing load, then processing complexity is reduced, but the pitch between microphones becomes further narrow worsening separation capability
Solution Approach 1:
The patent changes the processing parameter from time-domain amplitude/phase analysis to frequency-domain phase difference calculation. This parameter change enables accurate sound-source separation with only two closely disposed microphones, eliminating the need to increase the number of microphones while avoiding the worsening of pitch between microphones.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
Sound-source separation method, apparatus and program which can emphasize or suppress and output sound coming from an arbitrary direction are provided with a little amount of calculation using microphones closely disposed to each other and without a special analysis. Filtering containing a delay by a specific time is performed on one of the pair of input signals which are input from microphones L, R. After the filtering, a pair of input signals InL and InR are alternately interchanged for each sampling by an interchanging circuit 2 to generate a pair of interchanged signals InA and InB. The one interchanged signal InB is multiplied by a coefficient m by an coefficient updating circuit 3 to generate an error signal of the interchanged signals InA and InB. The recurrence formula of the coefficient m containing the error signal is calculated to update the coefficient m for each sampling. The pair of input signals InL and InR are multiplied by the sequentially updated coefficient m and are output.