Hybrid noise suppression for communication systems
A method for hybrid noise suppression involves receiving a first processed audio signal and a second processed audio signal from an audio device. The first processed audio signal results from a comprehensive audio processing including a noise spectrum estimate-based noise suppression performed on a noisy audio input signal obtained by the audio device. The second processed signal results from a partial audio processing excluding the noise spectrum estimate-based noise suppression performed on the noisy audio input signal. The method further involves temporally aligning the second processed audio signal with the first processed audio signal, predicting a noise suppression parameter using a neural network model operating on the second processed audio signal after the temporal alignment, generating a noise-suppressed audio output signal from the first processed audio signal after the temporal alignment using the noise suppression parameter, and outputting the noise-suppressed audio output signal.
1. A method for hybrid noise suppression, comprising:
receiving a first processed audio signal and a second processed audio signal from an audio device,
wherein the first processed audio signal results from a comprehensive audio processing including a noise spectrum estimate-based noise suppression performed on a noisy audio input signal obtained by the audio device,
wherein the second processed signal results from a partial audio processing excluding the noise spectrum estimate-based noise suppression performed on the noisy audio input signal;
temporally aligning the second processed audio signal with the first processed audio signal;
predicting a noise suppression parameter using a neural network model operating on the second processed audio signal after the temporal alignment;
generating a noise-suppressed audio output signal from the first processed audio signal after the temporal alignment, using the noise suppression parameter; and
outputting the noise-suppressed audio output signal.
2. The method of claim 1 , wherein predicting the noise suppression parameter comprises:
transforming the second processed audio signal from a time domain into a frequency domain in a plurality of frequency sub-bands;
generating a feature vector from the second processed audio signal in the plurality of frequency sub-bands; and
obtaining a set of sub-band gain values by applying the neural network model to the feature vector, wherein the set of sub-band gain values forms the noise suppression parameter.
3. The method of claim 1 , wherein generating the noise-suppressed audio output signal comprises:
transforming the first processed audio signal from a time domain into a frequency domain in a plurality of frequency sub-bands;
scaling, according to sub-band gain values provided as the noise suppression parameter, the first processed audio signal in the plurality of frequency sub-bands to generate the noise-suppressed audio output signal; and
transforming the noise-suppressed audio output signal from the frequency domain to the time domain.
4. The method of claim 1 , wherein the neural network model is a deep neural network.
5. The method of claim 4 , wherein the deep neural network comprises gated recurrent units.
6. The method of claim 1 , wherein the comprehensive audio processing further comprises at least one selected from the group consisting of a beamforming, an automatic gain control, an equalizing, an echo cancellation, and a limiting.
7. A system for hybrid noise suppression, comprising a host device, memory; and
circuitry for performing operations using the memory, the operations comprising:
receiving a first processed audio signal and a second processed audio signal from an audio device,
wherein the first processed audio signal results from a comprehensive audio processing including a noise spectrum estimate-based noise suppression performed on a noisy audio input signal obtained by the audio device,
wherein the second processed signal results from a partial audio processing excluding the noise spectrum estimate-based noise suppression performed on the noisy audio input signal;
temporally aligning the second processed audio signal with the first processed audio signal;
predicting a noise suppression parameter using a neural network model operating on the second processed audio signal after the temporal alignment;
generating a noise-suppressed audio output signal from the first processed audio signal after the temporal alignment, using the noise suppression parameter; and
outputting the noise-suppressed audio output signal.
8. The system of claim 7 , wherein the host device is one selected from the group consisting of a smartphone, a tablet computer, and a personal computer.
9. The system of claim 7 , further comprising the audio device.
10. The system of claim 9 , wherein the audio device is a telecommunication headset.
11. The system of claim 10 , wherein the audio device comprises:
an audio source; and
a computing system configured to:
obtain a noisy audio input signal from the audio source;
perform the comprehensive audio processing to generate the first processed audio signal from the noisy audio input signal;
perform the partial audio processing to generate the second processed audio signal from the noisy audio input signal; and
transmit the first processed audio signal and the second processed audio signal to the host device.
12. The system of claim 11 , wherein the audio source is a microphone.
13. The system of claim 7 , wherein the host device comprises at least one selected from the group consisting of a Bluetooth interface and a USB interface to interface with the audio device.