IP Library › Granted Patent US 12694259
Granted Patent B2
US 12694259 · App. 17/448,345 · Granted Jul 28, 2026

Low power multi-stage selectable neural network suppression

Inventors: Mouna Elkhatib (Irvine, CA); Adil Benyassine (Irvine, CA); Aruna Vittal (Irvine, CA); Daniel Schoch (Irvine, CA); Eli Uc (Irvine, CA)
Assignee: Aondevices, Inc.
G06N3/04
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12694259
App. No.
17/448,345
Filed
Sep 21, 2021
Granted
Jul 28, 2026
Kind
B2
Art Unit
2144
USPC
706/22
Abstract

A multi-stage selectable neural network noise suppression system has a first stage noise pattern selection neural network receptive to an input signal. An automatic noise classification is generated based upon an evaluation of the input signal. A noise suppression weight table stores one or more sets of automatic noise suppression weight values corresponding to the generated automatic noise classifications. A second stage noise pattern suppression neural network then selectively applies a specific automatic targeted noise suppression based upon the automatic noise suppression weight values. This approach balances performance quality and power/memory usage that efficiently tailors noise detection and suppression.

Claims (51)

1 . An article of manufacture comprising a non-transitory program storage medium readable by a data processing apparatus, the medium tangibly embodying one or more programs of instructions executable by the data processing apparatus to perform a method for suppressing noise in a signal, comprising the steps of:

receiving an input signal from a signal source;

evaluating the input signal with a first stage noise pattern selection neural network, an automatic noise classification being generated from the evaluation of the input signal;

receiving an external input of a selection of one of multiple predefined noise classifications, each of the predefined noise classifications having a set of manual noise suppression weight values;

retrieving a set of automatic noise suppression weight values stored in a noise suppression weight table that correlates noise classifications to sets of noise suppression weight values, the retrieved set of the automatic noise suppression weight values corresponding to the automatic noise classification as generated by the first stage noise pattern selection neural network; and

selectively applying an automatic targeted noise suppression to the input signal with a second stage noise pattern suppression neural network trained and operating independently of the first stage noise pattern selection neural network, the automatic targeted noise suppression being based upon the retrieved set of automatic noise suppression weight values.

2 . The article of manufacture of claim 1 , wherein the method further includes:

selectively applying either the automatic targeted noise suppression to the input signal or a manual targeted noise suppression based upon the manual noise suppression weight values corresponding to the selected one of the predefined noise classifications.

3 . The article of manufacture of claim 1 , wherein the method further includes:

combining selected set of automatic noise suppression weight values and the manual noise suppression weight values into a set of combined noise suppression weight values; and

applying a combined targeted noise suppression to the input signal based upon the set of combined noise suppression weight values.

4 . The article of manufacture of claim 1 , wherein the external input is received with a mechanical switch.

5 . The article of manufacture of claim 1 , wherein the method further includes:

generating a graphical user interface on a display device showing a menu of options for the multiple predefined noise classifications;

wherein receiving the external input corresponds to an input provided to the graphical user interface translatable to a selection of one of the options of the displayed menu.

6 . The article of manufacture of claim 1 , wherein the input signal is selected from a group consisting of: an audio signal, an infrasound signal, a radio frequency signal, an electrical signal, and an image stream.

7 . A multi-stage selectable neural network noise suppression system, comprising:

a first stage noise pattern selection neural network receptive to an input signal and an automatic noise classification being generated based upon an evaluation of the input signal by the first stage noise pattern selection neural network;

a noise suppression weight table correlating noise classifications to specific sets of automatic noise suppression weight values;

a second stage noise pattern suppression neural network trained and operating independently of the first stage noise pattern selection neural network and receptive to the input signal, a specific automatic targeted noise suppression being selectively applied to the input signal based upon a selected one of the automatic noise suppression weight values provided to the second stage noise pattern suppression neural network as retrieved from the noise suppression weight table with the automatic noise classification generated by the first stage noise pattern selection neural network; and

a manual noise suppression pattern selector receptive to an external input of a selection of one of multiple predefined noise classifications each with manual noise suppression weight values;

wherein the second stage noise pattern suppression neural network selectively applies a manual targeted noise suppression based upon the manual noise suppression weight values associated with the selected one of the predefined noise classifications.

8 . The multi-stage selectable neural network noise suppression system of claim 7 , wherein the second stage noise pattern suppression neural network applying the automatic targeted noise suppression is supplemental to the application of the manual targeted noise suppression, the automatic noise suppression weight values corresponding to the automatic noise classification generated by the first stage noise pattern selection neural network are combined with the manual noise suppression weight values associated with the selected one of the predefined noise classifications.

9 . The multi-stage selectable neural network noise suppression system of claim 7 , wherein the external input is a mechanical switch.

10 . The multi-stage selectable neural network noise suppression system of claim 7 , further comprising a graphical user interface with a display showing a menu of selectable options for the multiple predefined noise classifications and the external input being translatable to an invocation of one of the selectable options.

11 . The multi-stage selectable neural network noise suppression system of claim 7 , wherein the input signal is selected from a group consisting of: an audio signal, an infrasound signal, a radio frequency signal, an electrical signal, and an image stream.

12 . The multi-stage selectable neural network noise suppression system of claim 7 , further comprising:

a data processor implementing the first stage noise pattern selection neural network and the second stage noise pattern suppression neural network; and

a memory storing the noise suppression weight table.

13 . The multi-stage selectable neural network noise suppression system of claim 12 , further comprising:

an input audio transducer receptive to an audio wave, the audio wave being converted to the input signal.

14 . An article of manufacture comprising a non-transitory program storage medium readable by a data processing apparatus, the medium tangibly embodying one or more programs of instructions executable by the data processing apparatus to perform a method for suppressing noise in a signal, the method comprising the steps of:

receiving an input signal from a signal source;

evaluating the input signal with a first stage noise pattern selection neural network, an automatic noise classification being generated from the evaluation of the input signal;

retrieving a set of automatic noise suppression weight values stored in a noise suppression weight table that correlates noise classifications to sets of noise suppression weight values, the retrieved set of the automatic noise suppression weight values corresponding to the automatic noise classification as generated by the first stage noise pattern selection neural network;

selectively applying an automatic targeted noise suppression to the input signal with a second stage noise pattern suppression neural network trained and operating independently of the first stage noise pattern selection neural network, the automatic targeted noise suppression being based upon the retrieved set of automatic noise suppression weight values;

detecting a noise profile change in the input signal;

evaluating the input signal with the first stage noise pattern selection neural network, a secondary automatic noise classification being generated therefrom; and

updating the set of automatic noise suppression values correlated to the secondary automatic noise classification while the automatic targeted noise suppression is being applied to the input signal.

15 . The article of manufacture of claim 14 , wherein the method further includes:

receiving an external input of a selection of one of multiple predefined noise classifications, each of the predefined noise classifications having a set of manual noise suppression weight values.

16 . The article of manufacture of claim 15 , wherein the method further includes:

selectively applying either the automatic targeted noise suppression to the input signal or a manual targeted noise suppression based upon the manual noise suppression weight values corresponding to the selected one of the predefined noise classifications.

17 . The article of manufacture of claim 15 , wherein the method further includes:

combining selected set of automatic noise suppression weight values and the manual noise suppression weight values into a set of combined noise suppression weight values; and

applying a combined targeted noise suppression to the input signal based upon the set of combined noise suppression weight values.

18 . The article of manufacture of claim 15 , wherein the method further includes:

generating a graphical user interface on a display device showing a menu of options for the multiple predefined noise classifications;

wherein receiving the external input corresponds to an input provided to the graphical user interface translatable to a selection of one of the options of the displayed menu.

19 . The article of manufacture of claim 15 , wherein the external input is received with a mechanical switch.

20 . The article of manufacture of claim 14 , wherein the input signal is selected from a group consisting of: an audio signal, an infrasound signal, a radio frequency signal, an electrical signal, and an image stream.