Sub-vocalization Neural Signal Mapping via Artificial Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current sub-vocalization technologies face limitations in accurately translating neural signals into audio communications, particularly in noisy environments and require prior feature classification, which hampers their effectiveness in real-time applications.
Innovation Solution
The method employs artificial neural networks (ANNs) for continuous-to-continuous mapping of neural signal data from body sensors, such as EEG, EMG, and FNIRS, to create a common feature space that links neural signals to audio communications without prior classification, enabling accurate sub-vocalization speech reproduction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If prior feature classification is used to process neural signal data, then the processing structure is simpler and more interpretable, but the system cannot capture continuous nonlinear relationships and requires extensive manual feature engineering
Solution Approach 1:
The patent replaces traditional mechanical feature classification and extraction systems with an artificial neural network that performs continuous nonlinear mapping. The ANN automatically learns relevant features from raw neural signal data through iterative training, eliminating the need for manual feature engineering while capturing complex continuous relationships between neural signals and speech outputs.
2Object-affected harmful factors
If noise cancelation is used in speech recognition, then some environmental noise is reduced, but high levels of environmental noise distortions and variations in noise levels remain ineffective
Solution Approach 1:
The patent introduces an artificial neural network as an intermediary between neural signal acquisition and speech output. The ANN learns to map neural signals to speech outputs through iterative training, effectively bypassing environmental noise issues entirely by operating at the neural signal level rather than the acoustic signal level, thus achieving reliable communication even in high-noise environments.
3Measurement precision
If iterative closed loop training is performed to generate large data sets, then the accuracy of sub-vocalization speech reproduction is improved, but the training time and computational resources increase
Solution Approach 1:
The patent performs preliminary iterative closed-loop training to generate comprehensive training data sets and establish the neural network mappings before actual sub-vocalization communication is needed. This preliminary action allows the system to learn accurate mappings offline, enabling fast and accurate real-time speech reproduction without requiring extensive training during actual use.
Data Source
AI summary
Methods, systems and apparatuses are provided to perform a continuous-to-continuous mapping of neural signal data received from one or more body sensors connected to an user wherein the one or more body sensors monitors at least neural activities of the user of a sub-vocalized voice at a sensory level and sends the neural signal data to a processor. The processor receives the neural signal data in an iterative closed loop to train the processor and to generate a sufficiently large data set in the neural signal domain to link to a produced voice domain. The processor constructs a common feature space which associates the neural signal domain with the produced voice domain wherein the common feature space implicitly extracts features related to audio communications for linking neural signal domain data to the produced voice data without requiring any prior feature classification of the received neural signal data.