Singing voice separation with deep u-net convolutional networks

1. A method for training a neural network system, comprising: applying an audio signal to a U-Net neural network system, the audio signal including a vocal component and a non-vocal component, wherein applying the audio signal to the U-Net neural network includes converting the audio signal to an image;

Converting the image to an output audio signal and comparing the output audio signal of the U-Net neural network system to a target signal;

and adjusting at least one parameter of the U-Net neural network system to reduce a result of the comparing, for training the U-Net neural network system to estimate one of the vocal component and the non-vocal component.

2. The method of claim 1 , wherein the audio signal and the target signal represent different versions of a same musical song.

3. The method of claim 1 , wherein the target signal includes either vocal content or instrumental content.

4. The method of claim 1 , wherein the non-vocal component is an instrumental component, and the target signal represents an instrumental signal.

5. The method of claim 1 , wherein the non-vocal component is an instrumental component, and the target signal represents a vocal signal.

6. The method of claim 1 , further comprising obtaining the target signal by removing an instrumental component from a signal that includes vocal and instrumental components.

7. The method of claim 1 , further comprising identifying the audio signal and the target signal as a pair, wherein the identifying includes determining at least one of: (a) that the audio signal and the target signal relate to a same artist, (b) that a title associated with at least one of the audio signal and the target signal does not include predetermined information, and (c) that durations of the audio signal and the target signal differ by no more than a predetermined length of time.

8. The method of Claim 1 , wherein the U-Net comprises: a convolution path for encoding the image; and a deconvolution path for decoding the image encoded by the convolution path.

9. The method of claim 8 , further comprising applying an output of the deconvolution path as a mask to the image.

10. A method for estimating a component of a provided audio signal, comprising: converting the provided audio signal to an image; applying the image to a U-Net trained to estimate one of vocal content and instrumental content; and converting an output of the U-Net to an output audio signal, the output audio signal representing an estimate of either a vocal component of the provided audio signal or an instrumental component of the provided audio signal, depending on whether the U-Net is trained to estimate the vocal content or the instrumental content, respectively.

11. The method of claim 10 , wherein the U-Net comprises: a convolution path for encoding the image; and a deconvolution path for decoding the image encoded by the convolution path.

12. The method of claim 11 , further comprising applying an output of the deconvolution path as a mask to the image.

13. A system, comprising: a memory storing a program; and a processor, controllable by the program to perform a method for training a U-Net neural network system, the method including: applying an audio signal to the U-Net neural network system, the audio signal including a vocal component and a non-vocal component, wherein applying the audio signal to the U-Net neural network includes converting the audio signal to an image;

Converting the image to an output audio signal and comparing the output audio signal of the U-Net neural network system to a target signal;

and adjusting at least one parameter of the neural network system to reduce a result of the comparing, for training the U-Net neural network system to estimate one of the vocal component and the non-vocal component.

14. The system of claim 13 , wherein the non-vocal component is an instrumental component, and the target signal represents one of an instrumental signal or a vocal signal.

15. The system of claim 13 , wherein the U-Net neural network system comprises a U-Net including: a convolution path for encoding the image; and a deconvolution path for decoding the image encoded by the convolution path.

16. The system of claim 15 , wherein the method further comprises applying an output of the deconvolution path as a mask to the image.

17. A system, comprising: a memory storing a program; and a processor, controllable by the program to perform a method for estimating a component of a provided audio signal, the method comprising: converting the provided audio signal to an image; applying the image to a U-Net trained to estimate one of vocal content and instrumental content; and converting an output of the U-Net to an output audio signal, the output audio signal representing an estimate of either a vocal component of the provided audio signal or an instrumental component of the provided audio signal, depending on whether the U-Net is trained to estimate the vocal content or the instrumental content, respectively.

18. The system of claim 17 , where in the U-Net comprises: a convolution path for encoding the image; a deconvolution path for decoding the image encoded by the convolution path; and at least one concatenation between the convolution path and the deconvolution path.

19. The method according to claim 1 , wherein the U-Net neural network system has a plurality of skip connections between a plurality of layers of the U-Net neural network system.

20. The method according to claim 10 , wherein the U-Net has a plurality of skip connections between a plurality of layers of the U-Net.

21. The system according to claim 13 , wherein the U-Net neural network system has a plurality of skip connections between a plurality of layers of the U-Net neural network system.

22. The system according to claim 17 , wherein the U-Net has a plurality of skip connections between a plurality of layers of the U-Net.