Method and apparatus for estimating variability of background noise for noise suppression

A system comprising:

• one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
• receiving audio data that includes speech and background noise;
• processing the background noise included in the audio data to:
• obtain smoothed channel noise energy associated with the background noise;
• and
• determine a noise level of the background noise;
• determining whether a magnitude of the noise level of the background noise satisfies a noise level threshold;
• when the magnitude of the noise level of the background noise satisfies the noise level threshold, setting a bias energy level to a predetermined value based on the magnitude of the noise level of the background noise;
• adding the bias energy level set to the predetermined value to the smoothed channel noise energy to generate modified smoothed channel noise energy;
• determining, using the modified smoothed channel noise energy, a degree of variability of the background noise included in the audio data;
• determining that the degree of variability of the background noise included in the audio data satisfies a threshold;
• based on determining that the degree of variability of the background noise included in the audio data satisfies the threshold, selecting, from among at least a first noise suppressor that requires a first amount of processing power and a second noise suppressor that requires a second, greater amount of processing power, the second noise suppressor that requires the second, greater amount of processing power;
• based on selecting the second noise suppressor that requires the second, greater amount of processing power, processing the audio data using the second noise suppressor that requires the second, greater amount of processing power;
• and
• after the second noise suppressor that requires the second, greater amount of processing power has processed the audio data, obtaining, by performing speech recognition, a transcription of one or more of the portions of the audio data that include speech.

A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform operations comprising:

• receiving audio data that includes speech and background noise;
• processing the background noise included in the audio data to:
• obtain smoothed channel noise energy associated with the background noise;
• and
• determine a noise level of the background noise;
• determining whether a magnitude of the noise level of the background noise satisfies a noise level threshold;
• when the magnitude of the noise level of the background noise satisfies the noise level threshold, setting a bias energy level to a predetermined value based on the magnitude of the noise level of the background noise;
• adding the bias energy level set to the predetermined value to the smoothed channel noise energy to generate modified smoothed channel noise energy;
• determining, using the modified smoothed channel noise energy, a degree of variability of the background noise included in the audio data;
• determining that the degree of variability of the background noise included in the audio data satisfies a threshold;
• based on determining that the degree of variability of the background noise included in the audio data satisfies the threshold, selecting, from among at least a first noise suppressor that requires a first amount of processing power and a second noise suppressor that requires a second, greater amount of processing power, the second noise suppressor that requires the second, greater amount of processing power;
• based on selecting the second noise suppressor that requires the second, greater amount of processing power, processing the audio data using the second noise suppressor that requires the second, greater amount of processing power;
• and
• after the second noise suppressor that requires the second, greater amount of processing power has processed the audio data, obtaining, by performing speech recognition, a transcription of one or more of the portions of the audio data that include speech.