Mixture model based soft-clipping detection

Patent metadata

Field	Value
Publication number	US-10110187-B1
Application number	US-201715632498-A
Country	US
Kind code	B1
Filing date	Jun 26, 2017
Priority date	Jun 26, 2017
Publication date	Oct 23, 2018
Grant date	Oct 23, 2018

Abstract

Mixture model based soft-clipping detection includes receiving input audio samples, generating soft-clipping information indicating whether the input audio samples include soft-clipping distortion, and outputting the soft-clipping information. Generating the soft-clipping information includes fitting a mixture model to the input audio samples, wherein fitting the mixture model to the input audio samples includes generating a fitted mixture model, such that the fitted mixture model has fitted parameters, and evaluating a soft-clipping distortion metric based on the parameters of the fitted mixture model, wherein evaluating the soft-clipping distortion metric includes identifying a soft-clipping distortion value.

First claim

What is claimed is:

1. A method comprising: receiving input audio samples; generating soft-clipping information indicating whether the input audio samples include soft-clipping distortion, wherein generating the soft-clipping information includes: fitting a mixture model to the input audio samples, wherein fitting the mixture model to the input audio samples includes generating a fitted mixture model, such that the fitted mixture model has fitted parameters; and evaluating a soft-clipping distortion metric based on the parameters of the fitted mixture model, wherein evaluating the soft-clipping distortion metric includes identifying a soft-clipping distortion value; and outputting the soft-clipping information; wherein the fitted mixture model includes: a Laplacian distribution having a zero mean; a first Gaussian distribution; and a second Gaussian distribution, wherein the first Gaussian distribution and the second distribution are symmetrical.

2. The method of claim 1 , wherein evaluating the soft-clipping distortion metric includes:
evaluating the soft-clipping distortion metric using a trained machine learning model.

3. The method of claim 1 , wherein generating the soft-clipping information includes:
identifying a sequence of temporal portions of the input audio samples; and
fitting the mixture model to the input audio samples by:
for each temporal portion from the sequence of temporal portions:
fitting the mixture model to the respective temporal portion; and
generating respective soft-clipping information for each respective temporal portion.

4. A method comprising: receiving an input data signal, wherein the input data signal includes input audio samples; generating soft-clipping information indicating whether the input audio samples include soft-clipping distortion, wherein generating the soft-clipping information includes: fitting a mixture model to the input audio samples; and outputting the soft-clipping information; wherein the mixture model is a restricted mixture model; wherein the restricted mixture model includes: a Laplacian distribution having a zero mean; a first Gaussian distribution; and a second Gaussian distribution, wherein the first Gaussian distribution and the second Gaussian distribution are symmetrical.

5. The method of claim 4 , wherein fitting the mixture model to the input audio samples includes:
fitting the mixture model to the input audio samples using expectation maximization.

6. The method of claim 4 , wherein fitting the mixture model to the input audio samples includes:
generating a fitted mixture model, such that parameters of the fitted mixture model include:
a weight of a Gaussian distribution of the fitted mixture model;
a mean of the Gaussian distribution of the fitted mixture model;
a standard deviation of the Gaussian distribution of the fitted mixture model; and
a standard deviation of a Laplacian distribution of the fitted mixture model.

7. The method of claim 6 , wherein generating the soft-clipping information includes:
evaluating a soft-clipping distortion metric based on the parameters of the fitted mixture model.

8. The method of claim 7 , wherein evaluating the soft-clipping distortion metric includes:
generating an indication that the input audio samples include soft-clipping distortion in response to a determination that the mean of the Gaussian distribution of the fitted mixture model exceeds a first multiple of the standard deviation of the Gaussian distribution of the fitted mixture model and that the mean of the Gaussian distribution of the fitted mixture model exceeds a second multiple of the standard deviation of the Laplacian distribution of the fitted mixture model.

9. The method of claim 8 , wherein the first multiple is three and the second multiple is two.

10. The method of claim 7 , wherein evaluating the soft-clipping distortion metric includes:
identifying a soft-clipping distortion value indicating a severity of the soft-clipping distortion.

11. The method of claim 10 , wherein identifying the soft-clipping distortion value includes:
identifying a third multiple of the weight of the Gaussian distribution of the fitted mixture model as the soft-clipping distortion value.

12. The method of claim 11 , wherein the third multiple is two.

13. The method of claim 10 , wherein identifying the soft-clipping distortion value includes:
identifying, as the soft-clipping distortion value, a result of evaluating the parameters of the fitted mixture model using a machine learning model.

14. The method of claim 4 , wherein fitting the mixture model to the input audio samples includes:
identifying a defined cardinality of input audio quantiles from the input audio samples; and
fitting the mixture model to the input audio quantiles.

15. The method of claim 4 , wherein fitting the mixture model to the input audio samples includes:
identifying hard-clipped audio samples from the input audio samples;
identifying background audio samples from the input audio samples;
identifying target audio samples including the input audio samples other than the hard-clipped audio samples and the background audio samples; and
fitting the mixture model to the target audio samples.

16. The method of claim 4 , wherein generating the soft-clipping information includes:
identifying a sequence of temporal portions of the input audio samples; and
fitting the mixture model to the input audio samples by:
for each temporal portion from the sequence of temporal portions:
fitting the mixture model to the respective temporal portion; and
generating soft-clipping information indicating whether the respective temporal portion includes soft-clipping distortion.

17. The method of claim 4 , wherein the input data signal is an audio signal or an audio signal component of a video signal.

18. A method comprising:
receiving an input data signal, wherein the input data signal includes input audio samples;
generating soft-clipping information for the input data signal, wherein generating the soft-clipping information includes:
identifying a mixture model including:
a Laplacian distribution having a zero mean;
a first Gaussian distribution; and
a second Gaussian distribution, wherein the first Gaussian distribution and the second Gaussian distribution are symmetrical;

identifying a sequence of temporal portions of the input audio samples;
for each temporal portion from the sequence of temporal portions:
generating a respective fitted mixture model by fitting the mixture model to the respective input audio samples from the temporal portion using expectation maximization;
generating respective soft-clipping information indicating whether the temporal portion includes soft-clipping distortion by evaluating a soft-clipping distortion metric based on parameters of the respective fitted mixture model, wherein, in response to a determination that the temporal portion includes soft-clipping distortion, generating the respective soft-clipping information includes identifying a respective soft-clipping distortion value indicating a severity of the soft-clipping distortion for the respective input audio samples from the temporal portion; and
including the respective soft-clipping information for the temporal portion in the soft-clipping information for the input data signal;

generating an average soft-clipping distortion value for the input data signal;
including the average soft-clipping distortion value for the input data signal in the soft-clipping information for the input data signal;

Assignees

• Google Llc

Inventors

• Luebs Alejandro
• Obermeyer Fritz

Classifications

• G06F2218/02

  Preprocessing · CPC title

• H03G7/007 Primary

  of digital or coded signals · CPC title

• H04L1/20

  using signal quality detector · CPC title

• H03K5/08 Primary

  by limiting; by thresholding; by slicing, i.e. combined limiting and thresholding (H03K5/07 takes precedence; comparing one pulse with another H03K5/22; providing a determined threshold for switching H03K17/30) · CPC title

• G10L25/69

  for evaluating synthetic or decoded voice signals · CPC title