Spiking Neural Network Auditory Source Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in effectively separating mixed audio signals into their individual auditory sources, particularly in monaural, unsupervised, and online auditory source segregation for applications like speech enhancement and activity detection.
Innovation Solution
A spiking neural network-based method that selects an audio attribute, represents it as a spiking event, and determines coincidence with a single source by processing audio signals through a network of neurons with synaptic connections and memristor elements, utilizing spike-timing-dependent plasticity for adaptive learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional computational techniques are used for auditory source separation, then the system can process audio signals, but the complexity of the task and data makes the design of the function burdensome and impractical
Solution Approach 1:
The spiking neural network performs unsupervised learning, allowing the system to automatically infer source separation functions from audio observations without requiring manual function design or training data. The network self-organizes to separate mixed audio sources based on temporal coherence principles, eliminating the need for complex pre-programmed separation algorithms.
2Ease of operation
If conventional audio processing methods are used, then processing can be performed, but it is cumbersome and inadequate for monaural unsupervised source segregation
Solution Approach 1:
The patent replaces conventional mechanical/audio processing methods with a biologically-inspired spiking neural network that operates on temporal coincidence detection. Instead of using traditional signal processing algorithms, the system uses networks of spiking neurons that naturally perform source separation through their temporal response properties, simplifying the operational complexity.
3Adaptability or versatility
If prior knowledge of target source is required, then source separation can be achieved, but it limits applicability to unsupervised scenarios
Solution Approach 1:
The spiking neural network is pre-configured with temporal coincidence detection capabilities and connectivity structures that enable it to automatically perform source separation when presented with mixed audio signals. This preliminary structural configuration allows the network to achieve accurate source identification without requiring prior knowledge or training on specific target sources.
Data Source
AI summary
A method of audio source segregation includes selecting an audio attribute of an audio signal. The method also includes representing a portion of the audio attribute that is dominated by a single source as a source spiking event. In addition, the method includes representing a remaining portion of the audio signal as an audio signal spiking event. The method further includes determining whether the remaining portion coincides with the single source based on coincidence of the source spiking event and audio signal spiking event.