Wind noise detection and suppression
View Patent ↗Wind noise is detected in and removed from an acoustic signal. Features may be extracted from the acoustic signal. The extracted features may be processed to classify the signal as including wind noise or not. The wind noise may be removed before or during processing of the acoustic signal. The wind noise may be suppressed by estimating a wind noise model, deriving a modification, and applying the modification to the acoustic signal. In audio devices with multiple microphones, the channel exhibiting wind noise (i.e., acoustic signal frame associated with the wind noise) may be discarded for the frame in which wind noise is detected.
1. A method for performing noise reduction, comprising:
transforming an acoustic signal from time domain to frequency domain sub-band signals, the acoustic signal representing at least one captured sound;
extracting, using at least one hardware processor, a feature from a sub-band of the transformed acoustic signal;
detecting the presence of wind noise based on the feature;
generating a modification to suppress the wind noise based on the feature; and
before reducing other noise within the transformed acoustic signal, applying the modification to suppress the wind noise.
2. The method of claim 1 , wherein the feature includes a ratio between an energy level in a low frequency sub-band and a total signal energy.
3. The method of claim 1 , wherein the feature includes a variance of a ratio between an energy in a low frequency sub-band and a total signal energy.
4. The method of claim 1 , further comprising characterizing at least one of the sub-band signals as having wind noise.
5. The method of claim 4 , wherein the characterizing is based on a characterization engine trained with wind noise data.
6. The method of claim 5 , wherein an output of the characterization engine includes a binary classification.
7. The method of claim 4 , further comprising smoothing the characterization of wind noise over frames of the transformed acoustic signal.
8. The method of claim 1 , wherein the modification includes deriving a wind noise model by fitting a function to a signal spectrum for the transformed acoustic signal.
9. The method of claim 1 , further comprising:
extracting another feature from the sub-band of the transformed acoustic signal; and
detecting the presence of wind noise further based on the other feature.
10. The method of claim 9 , wherein the feature and the other feature include at least two of: a ratio between energy levels in low frequency sub-bands and a total signal energy, a mean of the ratio, a variance of the ratio, and a coherence between microphone signals.
11. The method of claim 1 , the other noise being environmental noise other than wind noise.
12. A system for reducing noise in an acoustic signal, the system comprising:
a wind noise characterization engine executable, using at least one hardware processor, to provide a wind noise characterization of a first acoustic signal, the first acoustic signal representing at least one captured sound;
a mask generator executable to generate a modification to suppress wind noise; and
a modifier module configured to apply the modification to suppress the wind noise based on the wind noise characterization, before environmental noise is reduced within the first acoustic signal.
13. The system of claim 12 , further comprising a memory and a first microphone configured to receive the first acoustic signal.
14. The system of claim 12 , further comprising a feature extraction module to extract features from the first acoustic signal, the wind noise characterization based on the features.
15. The system of claim 12 , further comprising a transform module to transform the first acoustic signal from a time domain to a frequency domain.
16. The system of claim 12 , further comprising a second acoustic signal, the second acoustic signal representing at least one captured sound, and the wind noise characterization engine configured to characterize the first and second acoustic signals independently.
17. The system of claim 16 , further comprising determining a coherence function between the first and second acoustic signals.
18. A non-transitory computer readable storage medium having embodied thereon a program, the program being executable by a processor to perform a method for reducing noise in an audio signal, the method comprising:
transforming an acoustic signal from time domain to frequency domain sub-band signals, the acoustic signal representing at least one captured sound;
extracting, using at least one hardware processor, a feature from a sub-band of the transformed acoustic signal;
detecting the presence of wind noise based on the feature;
generating a modification to suppress the wind noise based on the feature; and
before reducing environmental noise within the transformed acoustic signal, applying the modification to suppress the wind noise.
19. The non-transitory computer readable storage medium of claim 18 , the method further comprising generating the modification to suppress the wind noise based on the feature.
20. The non-transitory computer readable storage medium of claim 19 , wherein the modification to suppress the wind noise comprises discarding at least one frame of the transformed acoustic signal, wherein the at least one frame exhibits the wind noise.