Methods and apparatus for selective microphone signal combining
View Patent ↗Methods and apparatus for frequency selective signal mixing for speech enhancement. In one embodiment frequency-based channel selection is performed for signal magnitude, signal energy, and noise estimate using speaker activity detection information, signal-to-noise ratio, and/or signal level, Frequency-based channel selection is performed for a dynamic spectral floor to adjust the noise estimate using speaker dominance information. Noise reduction is performed on the signal for the selected channel.
1. A method, comprising:
receiving a plurality of microphone signals;
for each channel in the microphone signals, transforming the microphone signals into the frequency subband domain;
performing frequency-based channel selection using speaker activity detection information for mixing, by a mixer, signal spectra, signal power, and/or noise power of the channels;
using noise power spectral density estimates and speaker dominance information to generate attenuation vectors; and
performing noise reduction on output signals of the mixer by using an attenuation vector selected from the generated attenuation vectors.
2. The method according to claim 1 , further including performing the frequency-based channel selection for a dynamic spectral floor to adjust a noise estimate for the noise reduction using speaker dominance information.
3. The method according to claim 1 , further including mixing phase information with the selected signal channel.
4. The method according to claim 1 , further including selecting the channel having the maximum signal-to-noise ratio of the microphone signals for each subband.
5. The method according to claim 1 , further including selecting the channel having the highest signal energy of the microphone signals for each subband.
6. The method according to claim 1 , further including performing the frequency-based channel selection for signal amplitude and selecting phase of an active speaker signal based on the speaker activity detection information.
7. The method according to claim 1 , further including performing the noise reduction using filter coefficients that are limited by a channel-independent frequency-selected spectral floor.
8. The method according to claim 1 , further including computing dynamic spectral floors from speaker dominance weights of background noise for an active speaker and a previously active speaker.
9. An apparatus, comprising:
a non-transitory computer-readable medium having stored instructions that enable a machine to:
receive a plurality of microphone signals;
for each channel in the microphone signals, transform the microphone signals into the frequency subband domain;
perform frequency-based channel selection using speaker activity detection information for mixing, by a mixer, signal spectra, signal power, and/or noise power of the channels;
use noise power spectral density estimates and speaker dominance information to generate attenuation vectors; and
perform noise reduction on the output signals of the mixer by using an attenuation vector selected from the generated attenuation vectors.
10. The apparatus according to claim 9 further including performing the frequency-based channel selection for a dynamic spectral floor to adjust a noise estimate using speaker dominance information.
11. The apparatus according to claim 9 , further including instructions to mix phase information with the selected signal channel.
12. The apparatus according to claim 9 , further including instructions to select the channel having the maximum signal-to-noise ratio of the microphone signals for each subband.
13. The apparatus according to claim 9 , further including instructions to select the channel having the highest signal energy of the microphone signals for each subband.
14. The apparatus according to claim 9 , further including instructions to perform the frequency-based channel selection for signal amplitude and selecting phase of an active speaker signal based on the speaker activity detection information.
15. The apparatus according to claim 9 , further including instructions to perform the noise reduction using filter coefficients that are limited by a channel-independent frequency- selected spectral floor.
16. The apparatus according to claim 9 , further including instructions to compute dynamic spectral floors from the speaker dominance weights of background noise for the active speaker and a previously active speaker.
17. A system, comprising:
a processor; and
a memory that contains instructions that are readable by the processor to cause the processor to perform operations of:
a receiver to receive a plurality of microphone signals;
a speaker activity detector to transform the microphone signals into the frequency subband domain for each channel in the microphone signals;
a selective signal selector to perform frequency-based channel selection using speaker activity detection information for mixing signal spectra, signal power, and/or noise power of the channels;
a noise power spectral density evaluator to generate attenuation vectors using noise power spectral density estimates and speaker dominance information; and
a noise reducer to reduce noise on the mixed signal spectra, signal power, and/or noise power signals for generating a noise-reduced output signal by using an attenuation vector selected from the generated attenuation vectors.
18. The system according to claim 17 , wherein the system performs the frequency-based channel selection for a dynamic spectral floor to adjust a noise estimate using speaker dominance information.
19. The system according to claim 17 , wherein the selective signal selector selects the channel having the maximum signal-to-noise ratio of the microphone signals for each subband.
20. The system according to claim 17 , wherein the noise reducer performs the noise reduction using filter coefficients that are limited by a channel-independent frequency-selected spectral floor.