Selective noise suppression using a neural network
Apparatuses, systems, and techniques to selectively suppress noise in a conference call. In at least one embodiment, one or more neural networks may be used to identify a source of a noise component and cause a notification to be sent, informing a user of the noise source to confirm or decline suppression of the noise component.
1 . A processor communicatively coupled to a memory storing a plurality of processor-executable instructions, the processor reading and executing the processor-executable instructions to:
receive, via a conference call application implemented on at least one user device, an audio signal including multiple audio components;
use one or more neural networks comprising a first neural network to identify a source of a first audio component from the multiple audio components received at the conferencing call application;
predict, by the one or more neural networks comprising a second neural network, a filtering recommendation for the first audio component based on the source of the first audio component and context information relating to a conference call that is ongoing with the conference call application; and
display a user interface element of a notification at a user interface of the conference call application, the notification identifying the source of the first audio component and the filtering recommendation to a user of the conferencing call.
2 . The processor of claim 1 , wherein the executing the processor-executable instructions causes generation of one or more audio characteristics corresponding to the audio component.
3 . The processor of claim 2 , wherein the executing the processor-executable instructions causes an automatic filtering of the audio component based at least in part one the one or more audio characteristics.
4 . The processor of claim 1 , wherein the notification is sent to the user interface of a user device that captures an audio input comprising the identified audio component.
5 . The processor of claim 1 , wherein the notification includes a user selectable widget via which a user input indicating whether the audio component is to be filtered is received.
6 . The processor of claim 5 , wherein the executing the processor-executable instructions causes a restoration of a filtered audio component into the conference call when the user input indicates that the audio component is not to be filtered.
7 . The processor of claim 1 , wherein the executing the processor-executable instructions uses the one or more neural networks further to predict whether the audio component is to be filtered based at least in part on the identified source of the audio component, user profile information of the user, and/or context information relating to the conferencing call.
8 . A system comprising:
one or more processors;
a memory storing a plurality of processor-executable instructions, the one or more processors reading and executing the processor-executable instructions to
receive, via a conference call application implemented on at least one user device, an audio signal including multiple audio components;
use one or more neural networks comprising a first neural network to identify a source of a first audio component from the multiple audio components received at the conferencing call application;
predict, by the one or more neural networks comprising a second neural network, a filtering recommendation for the first audio component based on the source of the first audio component and context information relating to a conference call that is ongoing with the conference call application; and
display a user interface element of a notification at a user interface of the conference call application, the notification identifying the source of the first audio component and the filtering recommendation to a user of the conferencing call.
9 . The system of claim 8 , wherein the executing the processor-executable instructions causes generation of one or more audio characteristics corresponding to the audio component.
10 . The system of claim 8 , wherein the executing the processor-executable instructions causes an automatic filtering of the audio component based at least in part one the one or more audio characteristics.
11 . The system of claim 8 , wherein the notification is sent to the user interface of a user device that captures an audio input comprising the identified audio component.
12 . The system of claim 8 , wherein the notification includes a user selectable widget via which a user input indicating whether the audio component is to be filtered is received.
13 . The system of claim 12 , wherein executing the processor-executable instructions causes a restoration of a filtered audio component into the conference call when the user input indicating that the audio component is not to be filtered.
14 . The system of claim 8 , wherein the executing the processor-executable instructions uses the one or more neural networks further to predict whether the audio component is to be filtered based at least in part on the identified source of the audio component, user profile information of the user, and/or context information relating to the conferencing call.
15 . A non-transitory machine-readable medium having stored thereon a set of processor-executable instructions, which if performed by one or more processors reading and executing the set of processor-executable instructions, cause the one or more processors to at least:
receive, via a conference call application implemented on at least one user device, an audio signal including multiple audio components;
use one or more neural networks comprising a first neural network to identify a source of a first audio component from the multiple audio components received at the conferencing call application;
predict, by the one or more neural networks comprising a second neural network, a filtering recommendation for the first audio component based on the source of the first audio component and context information relating to a conference call that is ongoing with the conference call application; and
display a user interface element of a notification at a user interface of the conference call application, the notification identifying the source of the first audio component and the filtering recommendation to a user of the conferencing call.
16 . The medium of claim 15 , wherein the set of instructions, when performed by the one or more processors, cause generation of one or more audio characteristics corresponding to the one or more audio components.
17 . The medium of claim 15 , wherein the set of instructions, when performed by the one or more processors, cause an automatic filtering of the audio component based at least in part one the one or more audio characteristics.
18 . The medium of claim 15 , wherein the notification is sent to the user interface of a user device that captures an audio input comprising the identified audio component.
19 . The processor of claim 1 , wherein the processor further executes the plurality of processor-executable instructions to train the second neural network, to predict the filtering recommendation for a given identified audio component, on a training dataset comprising a plurality of identified audio sources, and one or more of:
user profile information,
associated conference information,
user real-time location information, or
annotated labels indicating whether the identified audio sources have been suppressed.
20 . The system of claim 8 , wherein the one or more processors further execute the plurality of processor-executable instructions to train the second neural network, to predict the filtering recommendation for a given identified audio component, on a training dataset comprising a plurality of identified audio sources, and one or more of:
user profile information,
associated conference information,
user real-time location information, or
annotated labels indicating whether the identified audio sources have been suppressed.