Sound Source Direction Estimation Using Frequency Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for sound source direction estimation with two microphones struggle to achieve high accuracy when dealing with multiple mixed sound sources, as they are limited in their ability to distinguish and estimate the directions of multiple sound sources accurately.
Innovation Solution
A sound source direction estimation device that calculates inter-microphone phase differences and single sound source masks for each frequency band, using these features to input into a multi-label classifier to determine the direction of sound sources, allowing for the estimation of multiple sound sources with high accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If two microphones are used for sound source direction estimation, then device complexity is reduced, but measurement precision deteriorates when dealing with multiple mixed sound sources
Solution Approach 1:
The patent segments the frequency spectrum into multiple frequency bins and processes each bin independently. For each frequency bin, it calculates the inter-microphone phase difference and determines whether only one sound source exists at that frequency. This segmentation allows the system to handle multiple sound sources across different frequencies while maintaining the simplicity of using only two microphones.
Solution Approach 2:
The patent applies local quality by treating each frequency bin differently based on whether it contains a single sound source or multiple sound sources. When only one sound source exists at a particular frequency bin, the system uses that frequency information for direction estimation. This selective approach allows accurate direction estimation for multiple sound sources overall while maintaining the simplicity of the two-microphone configuration.
2Device complexity
If only one sound source direction is estimated per frequency bin, then processing complexity is reduced, but the ability to estimate multiple sound sources deteriorates
Solution Approach 1:
The patent introduces a new dimension of analysis by examining the frequency domain characteristics of sound sources. Instead of estimating direction for all sound sources simultaneously in the time domain, it processes frequency bins independently and uses the inter-microphone phase difference at each frequency to determine sound source directions. This dimensional shift enables multiple sound source estimation while maintaining simple processing per frequency bin.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The device effectively estimates the directions of multiple sound sources with high accuracy using the inter-microphone phase differences and single sound source masks, even with a limited number of microphones, by ignoring frequency bins with no sound or multiple sound sources, enabling precise sound source direction determination.
Implementation Method 1
calculates an inter-microphone phase difference for every frequency band in a microphone pair
Data Source
AI summary
In a case where two microphones are used, sound source direction estimation of a plurality of sound sources can be performed with high accuracy. For this purpose, an inter-microphone phase difference is calculated for every frequency band in a microphone pair including two microphones that are installed apart from each other by a predetermined distance. Furthermore, for every frequency band in the microphone pair, a single sound source mask indicating whether or not a component of the frequency band is a single sound source is calculated. Then, the calculated inter-microphone phase difference and the calculated single sound source mask are input as feature quantities to a multi-label classifier, and a direction label associated with a sound source direction is output to the feature quantities.