Audio noise determination using one or more neural networks
Apparatuses, systems, and techniques are presented to reduce noise in audio. In at least one embodiment, one or more neural networks are used to determine a noise signal in one or more speech signals.
1. A processor, comprising:
one or more circuits to concurrently use two or more different portions of one or more neural networks to identify one or more noise signals in two or more different features of one or more speech signals.
2. The processor of claim 1 , wherein the one or more circuits are further to generate an audio spectrogram corresponding to one or more features extracted from the one or more speech signals.
3. The processor of claim 2 , wherein the one or more circuits are further to provide the audio spectrogram as input to the one or more neural networks, wherein the one or more neural networks generate an audio mask corresponding to the noise signal identified in the one or more speech signals.
4. The processor of claim 3 , wherein the two or more different portions of the one or more neural networks include two parallel paths for determining patterns in the audio spectrogram, the two parallel paths including a first path with a sequence of convolutional layers and a second path with one or more gated recurrent unit (GRU) layers.
5. The processor of claim 4 , wherein the one or more circuits are further to concatenate the patterns determined by the two parallel paths and process those concatenated patterns using a sequence of GRU layers to identify important noise patterns in the one or more speech signals for use in generating the audio mask.
6. The processor of claim 5 , wherein the one or more circuits are further to invert the audio mask and apply the inverted audio mask to the audio spectrogram in order to generate an output audio signal with the noise signal removed from the one or more speech signals.
7. A system comprising:
one or more processors to concurrently use two or more different portions of one or more neural networks to identify one or more noise signals in two or more different features of one or more speech signals.
8. The system of claim 7 , wherein the one or more processors are further to generate an audio spectrogram corresponding to one or more features extracted from the one or more speech signals.
9. The system of claim 8 , wherein the one or more processors are further to provide the audio spectrogram as input to the one or more neural networks, wherein the one or more neural networks generate an audio mask corresponding to the noise signal identified in the one or more speech signals.
10. The system of claim 9 , wherein the two or more different portions of the one or more neural networks include two parallel paths for determining patterns in the audio spectrogram, the two parallel paths including a first path with a sequence of convolutional layers and a second path with a gated recurrent unit (GRU) layer.
11. The system of claim 10 , wherein the one or more processors are further to concatenate the patterns determined by the two parallel paths and process those concatenated patterns using a sequence of GRU layers to identify important noise patterns in the one or more speech signals for use in generating the audio mask.
12. The system of claim 11 , wherein the one or more processors are further to invert the audio mask and apply the inverted audio mask to the audio spectrogram in order to generate an output audio signal with the noise signal removed from the one or more speech signals.
13. A method comprising:
concurrently using two or more different portions of one or more neural networks to identify one or more noise signals in two or more different features of one or more speech signals.
14. The method of claim 13 , further comprising:
generating an audio spectrogram corresponding to one or more features extracted from the one or more speech signals.
15. The method of claim 14 , further comprising:
providing the audio spectrogram as input to the one or more neural networks, wherein the one or more neural networks generate an audio mask corresponding to the noise signal identified in the one or more speech signals.
16. The method of claim 13 , wherein the two or more different portions of the one or more neural networks include two parallel paths for determining patterns in the audio spectrogram, the two parallel paths including a first path with a sequence of convolutional layers and a second path with a gated recurrent unit (GRU) layer.
17. The method of claim 16 , wherein the one or more circuits are further to concatenate the patterns determined by the two parallel paths and process those concatenated patterns using a sequence of GRU layers to identify important noise patterns in the one or more speech signals for use in generating the audio mask.
18. The method of claim 17 , further comprising:
inverting the audio mask and apply the inverted audio mask to the audio spectrogram in order to generate an output audio signal with the noise signal removed from the one or more speech signals.
19. A non-transitory machine-readable medium having stored thereon a set of instructions, which if performed by one or more processors, cause the one or more processors to at least:
concurrently use two or more different portions of one or more neural networks to identify one or more noise signals in two or more different features of one or more speech signals.
20. The non-transitory machine-readable medium of claim 19 , wherein the instructions when executed further cause the one or more processors to:
generate an audio spectrogram corresponding to one or more features extracted from the one or more speech signals.
21. The non-transitory machine-readable medium of claim 20 , wherein the instructions when executed further cause the one or more processors to:
provide the audio spectrogram as input to the one or more neural networks, wherein the one or more neural networks generate an audio mask corresponding to the noise signal identified in the one or more speech signals.
22. The non-transitory machine-readable medium of claim 21 , wherein the two or more different portions of the one or more neural networks include two parallel paths for determining patterns in the audio spectrogram, the two parallel paths including a first path with a sequence of convolutional layers and a second path with a gated recurrent unit (GRU) layer.
23. The non-transitory machine-readable medium of claim 22 , wherein the one or more circuits are further to concatenate the patterns determined by the two parallel paths and process those concatenated patterns using a sequence of GRU layers to identify important noise patterns in the one or more speech signals for use in generating the audio mask.
24. The non-transitory machine-readable medium of claim 23 , wherein the instructions, if performed by the one or more processors, further cause the one or more processors to:
invert the audio mask and apply the inverted audio mask to the audio spectrogram in order to generate an output audio signal with the noise signal removed from the one or more speech signals.
25. An audio denoising system, comprising:
one or more processors to concurrently use two or more different portions of one or more neural networks to identify one or more noise signals in two or more different features of one or more speech signals; and
memory for storing network parameters for the one or more neural networks.
26. The audio denoising system of claim 25 , wherein the one or more processors are further to generate an audio spectrogram corresponding to one or more features extracted from the one or more speech signals.
27. The audio denoising system of claim 26 , wherein the one or more processors are further to provide the audio spectrogram as input to the one or more neural networks, wherein the one or more neural networks generate an audio mask corresponding to the noise signal identified in the one or more speech signals.
28. The audio denoising system of claim 27 , wherein the two or more different portions of the one or more neural networks include two parallel paths for determining patterns in the audio spectrogram, the two parallel paths including a first path with a sequence of convolutional layers and a second path with a gated recurrent unit (GRU) layer.
29. The audio denoising system of claim 28 , wherein the one or more circuits are further to concatenate the patterns determined by the two parallel paths and process those concatenated patterns using a sequence of GRU layers to identify important noise patterns in the one or more speech signals for use in generating the audio mask.
30. The audio denoising system of claim 29 , wherein the one or more processors are further to invert the audio mask and apply the inverted audio mask to the audio spectrogram in order to generate an output audio signal with the noise signal removed from the one or more speech signals.