Back to results
US 8,543,526 B2Granted
Systems and methods using neural networks to reduce noise in audio signals
AssigneeThe Intellisis Corporation
InventorDouglas A. Moore
Life of this patent
Estimating the term…
Abstract
Systems, methods, and computer program products are provided to provide noise reduction for an input signal using a neural network. A feed-forward set of neuron groups is provided to enhance neuron activity within a particular frequency band based on prior reception of activity within that frequency band, and also to attenuate surrounding frequency bands. A surround-inhibition set of neuron groups further attenuates activity surrounding the stimulated frequency band.
Claims21
A system comprising:
- a physical processor;
- and
- physical storage media,
- the physical processor and physical storage media being configured to implement:
- a filter configured to create a feature vector from an input signal, wherein elements of the feature vector correspond to individual frequency bands of the input signal, wherein one or more individual elements of the feature vector correspond to adjacent individual frequency bands of the input signal;
- a feed-forward set of neuron groups implemented using the physical processor and physical storage media, wherein the feed-forward set of neuron groups is configured to enhance activity within a stimulated frequency band and attenuate activity surrounding the stimulated frequency band, wherein the stimulated frequency band corresponds to a particular element of the feature vector, wherein a radius of individual frequency bands are adjacent to the stimulated frequency band, and wherein the individual frequency bands that are outside of the radius of individual frequency bands are attenuated, wherein the feed-forward set of neuron groups comprise a first and second neuron group, the first neuron group feeding into the second neuron group, wherein the feed-forward set of neuron groups operate on subsequent time-slices of the input signal, wherein the first neuron group operates on a first time slice, and wherein the second neuron group operates on a second time slice;
- and
- a surround-inhibition set of neuron groups configured to further attenuate activity surrounding the stimulated frequency band.
5 more elements
A computer-implemented method, the method being implemented in a computer system that includes a physical processor and physical storage media, the method comprising:
- creating a feature vector from an input signal, wherein elements of the feature vector correspond to individual frequency bands of the input signal, wherein one or more individual elements of the feature vector correspond to adjacent individual frequency bands of the input signal;
- enhancing activity within a stimulated frequency band and attenuating activity surrounding the stimulated frequency band within a feed-forward set of neuron groups, wherein the stimulated frequency band corresponds to a particular element of the feature vector, wherein a radius of individual attenuated frequency bands are adjacent to the stimulated frequency band, and wherein the individual frequency bands that are outside of the radius of frequency bands are attenuated;
- and
- further attenuating activity surrounding the stimulated frequency band within a surround-inhibition set of neuron groups,
- wherein the feed-forward set of neuron groups comprise a first and second neuron group, the first neuron group feeding into the second neuron group, the feed-forward set of neuron groups operating on subsequent time-slices of the input signal, the first neuron group operating on a first time slice, and the second neuron group operating on a second time slice.
2 more elements
A non-transient computer-readable storage medium having stored thereon computer-executable instructions that, when executed by a computing device, cause the computing device to perform a method comprising:
- creating a feature vector from an input signal, wherein elements of the feature vector correspond to individual frequency bands of the input signal, wherein one or more individual elements of the feature vector correspond to adjacent individual frequency bands of the input signal;
- enhancing activity within a stimulated frequency band and attenuating activity surrounding the stimulated frequency band within a feed-forward set of neuron groups, wherein the stimulated frequency band corresponds to a particular element of the feature vector, wherein a radius of individual attenuated frequency bands are adjacent to the stimulated frequency band, and wherein the individual frequency bands that are outside of the radius of frequency bands are attenuated;
- and
- further attenuating activity surrounding the stimulated frequency band within a surround-inhibition set of neuron groups,
- wherein the feed-forward set of neuron groups comprise a first and second neuron group, the first neuron group feeding into the second neuron group, the feed-forward set of neuron groups operating on subsequent time-slices of the input signal, the first neuron group operating on a first time slice, and the second neuron group operating on a second time slice.
2 more elements
Dependencies are inferred from the claim text.