Granted Patent B2

US 12,524,497 · App. 18/439,970 · Granted Jan 13, 2026

Inventor: Gurunandan Krishnan Gorumkonda (Kirkland, WA)

Assignee: Snap Inc.

G06F18/2148G06N3/04G06N3/08G10L21/14G10L25/30G06F2218/08G06F2218/12

Abstract

Claims (64)

1 . A method comprising:

processing a first output image, generated with a first convolutional neural network (CNN), using a second CNN to generate a clean loss, the clean loss indicating an amount of the first output image that is not attributed to a first accompaniment portion;

training the first CNN based on the clean loss;

receiving an input song comprising a vocal portion and a second accompaniment portion;

transforming the input song to an input image;

processing the input image using the first CNN to generate a second output image;

transforming the second output image to an output accompaniment, the output accompaniment comprising the second accompaniment portion of the input song;

storing first feature values determined in processing the first output image by the second CNN;

processing a modified ground truth output song with the second CNN;

storing second feature values determined in processing the modified ground truth output song by the second CNN;

determining a feature loss based on differences between the second feature values and the first feature values; and

training the second CNN further based on the feature loss.

2 . The method of claim 1 , wherein the input image representing frequencies and intensities of the input song.

3 . The method of claim 1 , wherein the clean loss is a first clean loss, and wherein the method further comprises:

training the second CNN using a ground truth input song paired with the modified ground truth output song, wherein the modified ground truth output song is the ground truth input song with added vocal portions and an estimate of a second clean loss for the modified ground truth output song.

4 . The method of claim 1 , wherein the input image has a pixel height and a pixel width, and wherein processing the output image using the second CNN further comprises:

determining a first layer of the first features values by convolving a plurality of kernels with the output image to determine feature values for the first layer, wherein the plurality of kernels have kernel pixel heights less than the pixel height of the input image and kernel pixel widths less than the pixel width of the input image, and wherein each kernel of the plurality of kernels generates a sublayer of the first layer.

5 . The method of claim 3 , wherein the estimate of the second clean loss is based on an amount of the added vocal portions in proportion to the ground truth output song.

6 . The method of claim 1 further comprising:

determining a loss between the output accompaniment and a ground truth output song, wherein the input song is a ground truth input song paired with the ground truth output song; and

training the first CNN further based on the determined loss.

7 . The method of claim 1 , wherein the second CNN is trained to determine the clean loss.

8 . The method of claim 1 , wherein the accompaniment portion of the input song is blank or the vocal portion of the input song is blank.

9 . The method of claim 1 wherein the first CNN does not comprise a maximum pooling layer.

10 . The method of claim 1 wherein the second CNN comprises multiple convolution layers, a maximum pooling layer, an up-conversion layer, and a fully-connected layer.

11 . The method of claim 1 , wherein training further comprises:

training, using stochastic gradient descent, the first CNN based on the clean loss.

12 . A system comprising:

one or more computer processors; and

one or more computer-readable mediums storing instructions that, when executed by the one or more computer processors, cause the system to perform operations comprising:

processing a first output image, generated with a first convolutional neural network (CNN), using a second CNN to generate a clean loss, the clean loss indicating an amount of the first output image that is not attributed to a first accompaniment portion;

training the first CNN based on the clean loss;

receiving an input song comprising a vocal portion and a second accompaniment portion;

transforming the input song to an input image;

processing the input image using the first CNN to generate a second output image;

transforming the second output image to an output accompaniment, the output accompaniment comprising the second accompaniment portion of the input song;

storing first feature values determined in processing the first output image by the second CNN;

processing a modified ground truth output song with the second CNN;

storing second feature values determined in processing the modified ground truth output song by the second CNN;

determining a feature loss based on differences between the second feature values and the first feature values; and

training the second CNN further based on the feature loss.

13 . The system of claim 12 , wherein the input image representing frequencies and intensities of the input song.

14 . The system of claim 12 , wherein the clean loss is a first clean loss, and wherein the operations further comprise:

training the second CNN using a ground truth input song paired with the modified ground truth output song, wherein the modified ground truth output song is the ground truth input song with added vocal portions and an estimate of a second clean loss for the modified ground truth output song.

15 . The system of claim 12 , wherein the input image has a pixel height and a pixel width, and wherein processing the output image using the second CNN further comprises:

determining a first layer of the first features values by convolving a plurality of kernels with the output image to determine feature values for the first layer, wherein the plurality of kernels have kernel pixel heights less than the pixel height of the input image and kernel pixel widths less than the pixel width of the input image, and wherein each kernel of the plurality of kernels generates a sublayer of the first layer.

16 . A non-transitory computer-readable storage medium including instructions that, when processed by a computer, configure the computer to perform operations comprising:

processing a first output image, generated with a first convolutional neural network (CNN), using a second CNN to generate a clean loss, the clean loss indicating an amount of the first output image that is not attributed to a first accompaniment portion;

training the first CNN based on the clean loss;

receiving an input song comprising a vocal portion and a second accompaniment portion;

transforming the input song to an input image;

processing the input image using the first CNN to generate a second output image;

transforming the second output image to an output accompaniment, the output accompaniment comprising the second accompaniment portion of the input song;

storing first feature values determined in processing the first output image by the second CNN;

processing a modified ground truth output song with the second CNN;

storing second feature values determined in processing the modified ground truth output song by the second CNN;

determining a feature loss based on differences between the second feature values and the first feature values; and

training the second CNN further based on the feature loss.

17 . The non-transitory computer-readable storage medium of claim 16 , wherein the input image representing frequencies and intensities of the input song.

18 . The non-transitory computer-readable storage medium of claim 16 , wherein the clean loss is a first clean loss, and wherein the operations further comprise:

training the second CNN using a ground truth input song paired with the modified ground truth output song, wherein the modified ground truth output song is the ground truth input song with added vocal portions and an estimate of a second clean loss for the modified ground truth output song.

19 . The non-transitory computer-readable storage medium of claim 18 , wherein the estimate of the second clean loss is based on an amount of the added vocal portions in proportion to the ground truth output song.

20 . The non-transitory computer-readable storage medium of claim 16 , wherein the input image has a pixel height and a pixel width, and wherein processing the output image using the second CNN further comprises:

determining a first layer of the first features values by convolving a plurality of kernels with the output image to determine feature values for the first layer, wherein the plurality of kernels have kernel pixel heights less than the pixel height of the input image and kernel pixel widths less than the pixel width of the input image, and wherein each kernel of the plurality of kernels generates a sublayer of the first layer.

Assignments (1)

Continuity (2)

References Cited (25)