Spectral estimation of room acoustic parameters

Patent metadata

Field	Value
Publication number	US-10403300-B2
Application number	US-201616084771-A
Country	US
Kind code	B2
Filing date	Mar 17, 2016
Priority date	Mar 17, 2016
Publication date	Sep 3, 2019
Grant date	Sep 3, 2019

Abstract

A system and associated method for estimating one or more reverberation parameters includes an adaptive, parametric, linear prediction filter having one or more power spectral density (PSD) estimates of signals of one or more channels as inputs. The prediction filter estimates at least one parameter related to reverberation time, such as T 60 , and can further estimate an additional parameter, such as Direct-to-Reverberant Ratio (DRR). The prediction filter may be adapted during a period of reverberation by minimizing a cost function. Adaptation can include using a gradient descent approach, which can operate according to a step size provided by an adaptation controller configured to determine the period of reverberation. One or more microphones can provide the signals. The reverberation parameters estimated can be applied to a reverberation suppressor, with an estimator that does not require a training phase and without relying on assumptions of the user's position relative to the microphones.

First claim

What is claimed is:

1. A method of estimating one or more reverberation parameters, the method comprising:
applying an adaptive, parametric, linear prediction filter configured to receive one or more power spectral density (PSD) estimates of signals of one or more channels as inputs, the prediction filter estimating at least one reverberation parameter related to reverberation time, the prediction filter being adapted during a period of reverberation by minimizing a cost function.

2. The method of claim 1 , further comprising adapting the prediction filter using a gradient descent approach.

3. The method of claim 2 , wherein the gradient descent approach operates according to a step size, the step size being provided by an adaptation controller.

4. The method of claim 1 , further comprising determining the period of reverberation by detecting that only reverberation is present in the signals.

5. The method of claim 4 , wherein the period of reverberation is determined by estimating a direct-to-reverberant ratio.

6. The method of claim 5 , wherein estimating the direct-to-reverberant ratio provides spectral information of the presence of direct sound components over reverberant sound components.

7. The method of claim 6 , wherein the direct and reverberant sound components are described by spectral coherence functions computed for at least two of the signals, the coherent parts of the coherence functions being indicative of direct sound components, the diffuse parts of the coherence functions being indicative of the reverberant sound components.

8. The method of claim 4 , wherein determining the period of reverberation includes using decay detection.

9. The method of claim 8 , wherein using decay detection includes detecting decaying signal components by comparing two smoothed PSD estimates, a first of the two smoothed PSD estimates corresponding to a current frame and a second of the two smoothed PSD estimates being delayed by a selected number of frames relative to the current frame.

10. The method of claim 1 , wherein the prediction filter is implemented as an infinite impulse response (IIR) filter.

11. The method of claim 10 , further including computing a gradient of the cost function according to a recursive prediction error principle.

12. The method of claim 1 , wherein the prediction filter is implemented as a finite impulse response (FIR) filter.

13. The method of claim 1 , wherein the prediction filter further estimates a reverberation parameter related to the ratio of power of direct sound and power of reverberation.

14. The method of claim 1 , wherein the cost function is at least one of a cost function that is independent of power of the signals and a logarithmic cost function.

15. The method of claim 1 , further comprising applying the at least one reverberation parameter estimated to a reverberation suppressor.

16. A system for estimating one or more reverberation parameters, the system comprising:
an adaptive, parametric, linear prediction filter having one or more power spectral density (PSD) estimates of signals of one or more channels as inputs, the prediction filter configured to estimate at least one reverberation parameter related to reverberation time, the prediction filter being adapted during a period of reverberation by minimizing a cost function.

17. The system of claim 16 , further comprising an adaptation controller configured to determine the period of reverberation.

18. The system of claim 17 , wherein the prediction filter is adapted using a gradient descent approach, the gradient descent approach operating according to a step size provided by the adaptation controller.

19. The system of claim 16 , further comprising an array of microphones to provide the signals.

20. The system of claim 16 , further comprising a reverberation suppressor configured to suppress reverberation in at least one of the signals based on the at least one reverberation parameter.

Assignees

• Nuance Communications Inc

Inventors

• Wolff Tobias
• Desiraju Naveen Kumar

Classifications

• G10L21/0208

  Noise filtering · CPC title

• G01H7/00 Primary

  Measuring reverberation time {; room acoustic measurements} · CPC title

• G10L21/0224 Primary

  Processing in the time domain · CPC title

• G10L2021/02082

  the noise being echo, reverberation of the speech · CPC title

• H04R3/005

  for combining the signals of two or more microphones (specially adapted for hearing aids H04R25/407) · CPC title