Single Microphone Direct-to-Reverberant Ratio Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the direct-to-reverberant ratio in hearing devices are cumbersome, require multiple microphones, and rely on assumptions about the sound field, making them unsuitable for real-world applications, especially in hearing devices that need to process sound signals with low computational cost and simplicity.
Innovation Solution
A method using a single microphone to estimate the direct-to-reverberant ratio by determining energy values in time frames, identifying acoustic onsets, and inputting these into a machine learning algorithm trained to calculate the direct-to-reverberant ratio, which is computationally efficient and independent of signal level and microphone directivity patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple microphones are used for DRR estimation, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The single microphone serves multiple functions: it captures both direct sound and reverberation, performs self-analysis through machine learning to separate these components, and generates the DRR estimate without requiring additional dedicated sensors. The microphone essentially analyzes its own output signal to extract reverberation characteristics.
Solution Approach 2:
The patent transforms the sound signal into different parameter representations (time-domain energy values, frequency-domain features, statistical parameters) and feeds these transformed parameters into the machine learning model. This parameter transformation enables the single microphone to provide sufficient information for accurate DRR estimation.
2Measurement precision
If complex algorithms are used for DRR estimation, then measurement precision is improved, but computational cost increases
Solution Approach 1:
The machine learning model is trained offline in advance with large datasets to learn the complex patterns between sound signal features and DRR values. During actual operation, the pre-trained model performs rapid inference by simply evaluating the learned parameters, avoiding the need for complex real-time calculations.
Solution Approach 2:
The patent replaces traditional signal processing methods (mechanical/mathematical algorithms for reverberation analysis) with a data-driven machine learning approach. The complex reasoning is shifted from runtime computation to offline training, enabling efficient real-time operation with lower computational energy consumption.
3Device complexity
If assumptions about sound field are made, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
Instead of relying on pre-defined sound field assumptions (isotropic, diffuse, etc.), the machine learning model learns the actual acoustic environment characteristics directly from the sound signal captured by the microphone. The system adapts to whatever sound field conditions exist without requiring them to match theoretical models.
Solution Approach 2:
The patent employs dynamic adaptation where the machine learning model can be retrained or fine-tuned for different acoustic environments. The system transitions from static assumptions to dynamic learning, allowing it to accurately estimate DRR in varying sound field conditions without requiring complex real-time analysis of environmental parameters.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A method for estimating a direct-to-reverberant ratio (38) of a sound signal (30) is proposed. The method comprises: determining a first energy value of a sound signal (30) for a first time frame; assigning to an onset value of the first time frame a positive value, if the difference of the first energy value of the first time frame and a second energy value of a preceding second time frame is greater than a threshold, and a zero value otherwise; and determining the direct-to-reverberant ratio (38) by providing an onset signal (42) comprising the onset value to a machine learning algorithm (44), which has been trained to determine the direct-to-reverberant ratio (38) based on said onset signal.