Spatial adaptation in multi-microphone sound capture

• US 10,154,342 B2
• Filed: 11/23/2016
• Issued: 12/11/2018
• Est. Priority Date: 02/10/2011
• Status: Active Grant

• Current Assignee: Dolby International AB (Dolby Laboratories Incorporated)
• Original Assignee: Dolby International AB (Dolby Laboratories Incorporated)
• Inventors: Samuelsson, Leif Jonas
• Primary Examiner(s): PATEL, YOGESHKUMAR G
• Application Number:
  US15/360,838
• Publication Number:
  US 20170078791A1
• Time in Patent Office: 748 Days
• Field of Search: None
• CPC Class Codes: G10L 21/0232 Processing in the frequency domain
• CPC Class Codes: H04R 2430/20 Processing of the output signals of the acoustic transducers of an array for obtaining a desired directivity characteristic H04R2203/12 takes precedence
• CPC Class Codes: H04R 29/006 Microphone matching
• CPC Class Codes: H04R 3/005 for combining the signals of two or more microphones specially adapted for hearing aids H04R25/407
• CPC Class Codes: H04R 3/04 for correcting frequency response

Abstract

A spatial adaptation system for multiple-microphone sound capture systems and methods thereof are described. A spatial adaptation system includes an inference and weight module configured to receive a inputs. The inputs based on two or more input signals captured by at least two microphones. The inference and weight module to determine one or more weight values base on at least one of the inputs. The spatial adaptation system also including a noise magnitude ratio update module coupled with the inference and weight module. The noise magnitude ratio update module to determine an updated noise target based on the one or more weight values from the inference and weight module.

Claims

1. A spatial adaptation method for providing long term symmetry among a plurality of microphones of a device, comprising:

calculating a magnitude ratio (MR) by computing a ratio between a first energy representing first group of one or more near-field microphones and a second energy representing a second group of one or more far-field microphones;

using the MR to provide microphone matching of large variations without any situational assumptions to lower manufacturing cost and complexities, wherein the microphone matching depends on a difference between a near-field signal and a far-field signal; and

in accordance with the microphone matching, adjusting the first group of one or more near-field microphones and the second group of one or more far-field microphones to have similar performance characteristics,

wherein the step of calculating the magnitude ratio further comprises;

calculating a minima follower and a maxima follower configured to track a minimum magnitude ratio and a maximum magnitude ratio over time, applied separately and independently in each frequency band of a plurality of frequency bands; and

calculating an average magnitude ratio based on the minima follower and the maxima follower,

wherein a buffer stores the minimum magnitude ratio and the maximum magnitude ratio over time, and

wherein the average magnitude ratio is used to provide the microphone matching.

2. The method of claim 1, further comprising the step of:

updating a microphone match table of the plurality of microphones during real-time use of the device.

3. The method of claim 1, wherein the spatial adaptation employs a Gaussian mixture model (GMM) based inference that is optimized for a source or voice dominated signal (clean voice or speech), and one model is optimized for noise dominated signals (noise) to classify a desired source.

4. The method of claim 1, wherein the spatial adaptation is not performed if self noise is detected.

5. The method of claim 4, wherein the method of detecting microphone self noise is implemented in each decision band of a plurality of decision bands.

6. The method of claim 4, wherein the self noise detection is based on aggregated features aggregated across frequencies.

7. The method of claim 6, wherein the aggregated features is selected from a group comprising:

a scalar aggregated of frame power;

a scalar aggregate of phase differences;

or

a scalar aggregate of coherences.

8. The method of claim 7, wherein self noise is detected if one or more of the following conditions are met:

if the scalar aggregated of frame power is less than a first predefined threshold;

or

if the scalar aggregated of frame power is less than a second predefined threshold that is greater than the first predefined threshold and the scalar aggregate of phase differences is greater than a third predefined threshold and the scalar aggregate of coherence is less than a fourth predefined threshold.

9. The method of claim 1, wherein the spatial adaptation method determines the maximum magnitude ratio to protect against interfering sources by comparing the magnitude ratio of a current frame with a threshold derived from an estimate of maximum ratio that is produced by a near-field talker.

10. The method of claim 1, further comprising a step of processing an output of a minima and maxima search configured to smooth or compensation for a minimum bias or a maximum bias.

11. The method of claim 3, wherein the GMM, for each frequency band is optimized offline to model distributions of the MR from near-field and far-field training data.

12. The method of claim 1, wherein the spatial adaptation method is based on an estimation of a ratio between energies of a rear and front signal instead of a ratio between a front and a rear signal.

13. The method of claim 5, wherein a decision to update a noise target or not is performed in each of the plurality of decision bands, and wherein the plurality of decision bands are not the same as the plurality of frequency bands in which a plurality of features are determined.

14. The method of claim 13, wherein the noise target is a long-term average of a magnitude ratio for noise, wherein the noise target is used to modify one or more received signals, and wherein the noise target is employed to match the one or more received signals.

15. The method of claim 1, wherein the spatial adaptation method maintains a variable to determine when to update one or more noise targets.

16. The method of claim 3, wherein if speech is found in a decision band, a noise target in one or more feature bands are associated with that decision band and is updated using a different set of weights than when noise is found in the decision band.

17. The method of claim 3, where the GMM is optimized offline on features selected from a group comprising the following features;

a magnitude ratio feature;

a phase difference feature;

a frame power feature;

a coherence feature;

or

a delta magnitude ratio feature;

a delta phase difference feature;

a delta frame power feature;

or

a delta coherence feature.

18. A non-transitory computer readable storage medium, storing software instructions, which when executed by one or more processors cause performance of the method as recited in claim 1.

19. A computing device comprising one or more processors and one or more storage media storing a set of instructions which, when executed by the one or more processors, cause performance of the method as recited in claim 1.

20. The method of claim 1, wherein the buffer is not updated when a time frame and frequency band contains no acoustic stimuli.