Methods and systems for determining speech presence probability, speech enhancement methods and systems, and headphones
The present disclosure provides a method and system for determining a speech presence probability, a speech enhancement method and system, and a headphone. The speech presence probability and a speech absence probability in an iteration operation may be corrected by comparing an entropy of the speech presence probability and an entropy of a speech absence probability, such that a faster convergence speed and better convergence results may be obtained, thereby improving accuracy of an estimation of the speech presence probability and an accuracy of an estimation of a noise spatial covariance matrix, and then improving a speech enhancement effect of a minimum variance distortionless response (MVDR).
1 . A method for determining a speech presence probability applied to M microphones arranged in a preset array, M being an integer greater than 1, and the method comprises:
obtaining microphone signals output by the M microphones, the microphone signals satisfying a first model or a second model of a Gaussian distribution, one of the first model and the second model being a speech presence model, and the other of the first model and the second model being a speech absence model;
constructing an objective function based on a maximum likelihood estimation (MLE) and an expectation maximization (EM) algorithm;
determining optimization parameters, the optimization parameters including a first spatial covariance matrix corresponding to the first model and a second spatial covariance matrix corresponding to the second model;
determining initial values of the optimization parameters;
performing, based on the objective function and the initial values of the optimization parameters, a plurality of iterations on the optimization parameters until the objective function converges by:
determining whether the speech presence model is the first model or the second model based on an entropy of a first probability that the microphone signals satisfy the first model and an entropy of a second probability that the microphone signals satisfy the second model in the plurality of iterations, the first probability being complementary to the second probability; and
in response to that the objective function converges, outputting the first probability and the second probability corresponding to convergence values of the optimization parameters, and
determining, based on a determination result of whether the speech presence model is the first model or the second model and one of the first probability and the second probability corresponding to the convergence values, a probability that the microphone signals satisfy the speech presence model as the speech presence probability of the microphone signals and outputting the speech presence probability.
2 . The method of claim 1 , wherein
a first variance of the Gaussian distribution corresponding to the first model includes a product of a first parameter and the first spatial covariance matrix; and
a second variance of the Gaussian distribution corresponding to the second model includes a product of a second parameter and the second spatial covariance matrix.
3 . The method of claim 2 , further comprising:
in response to that the objective function converges, outputting the convergence values of the optimization parameters.
4 . The method of claim 1 , wherein the determining whether the speech presence model is the first model or the second model based on the entropy of the first probability and the entropy of the second probability in the plurality of iterations includes:
in each iteration of the plurality of iterations, determining the entropy of the first probability and the entropy of the second probability, and determining whether the speech presence model is the first model or the second model by:
in response to determining that the entropy of the first probability is greater than the entropy of the second probability, determining that the speech presence model is the second model; or
in response to determining that the entropy of the first probability is less than the entropy of the second probability, determining that the speech presence model is the first model.
5 . The method of claim 1 , wherein the determining whether the speech presence model is the first model or the second model based on the entropy of the first probability and the entropy of the second probability in the plurality of iterations includes:
in a first iteration of the plurality of iterations, determining the entropy of the first probability and the entropy of the second probability, and determining whether the speech presence model is the first model or the second model by:
in response to determining that the entropy of the first probability is greater than the entropy of the second probability, determining that the speech presence model is the second model; or
in response to determining that the entropy of the first probability is less than the entropy of the second probability, determining that the speech presence model is the first model.
6 . The method of claim 5 , wherein the performing the plurality of iterations on the optimization parameters further includes, in each iteration of the plurality of iterations:
correcting the first probability and the second probability based on the entropy of the first probability and the entropy of the second probability by:
in response to determining that the first model is the speech presence model and the entropy of the first probability is greater than the entropy of the second probability, designating the second probability as a corrected first probability and designating the first probability as a corrected second probability; or
in response to determining that the second model is the speech presence model and the entropy of the second probability is greater than the entropy of the first probability, designating the second probability as a corrected first probability and designating the first probability as a corrected second probability; and
updating the optimization parameters based on the corrected first probability and the corrected second probability.
7 . The method of claim 1 , wherein the performing the plurality of iterations on the optimization parameters further includes, in each iteration of the plurality of iterations:
performing a reversible correction on the optimization parameters by:
in response to determining that the optimization parameters are irreversible, correcting the optimization parameters based on a deviation matrix, the deviation matrix including one of an identity matrix, or a random matrix obeying a normal distribution or a uniform distribution.
8 . The method of claim 1 , further comprising:
determining a noise spatial covariance matrix of the microphone signals based on the speech presence probability;
determining filter coefficients corresponding to the microphone signals based on a minimum variance distortionless response (MVDR) algorithm and the noise spatial covariance matrix; and
outputting a target audio signal by combining the microphone signals based on the filter coefficients.
9 . A headphone, comprising:
a microphone array including the M microphones arranged in the preset array; and
a computing device in communication connection to the microphone array during operation, and implements the method of claim 1 .
10 . The headphone of claim 9 , wherein the M microphones are linearly distributed, M is less than or equal to 5, and a distance between adjacent microphones among the M microphones is in a range of 20 mm-40 mm.
11 . The headphone of claim 9 , further comprising:
a first housing, wherein the microphone array is mounted on the first housing, and the first housing includes a first interface including a first magnetic device; and
a second housing, wherein the computing device is mounted on the second housing, and the second housing includes a second interface including a second magnetic device,
wherein an adsorption force between the first magnetic device and the second magnetic device enables a detachable connection between the first housing and the second housing.
12 . The headphone of claim 11 , wherein
the first housing further includes one or more contact points disposed at the first interface and in communication connection to the microphone array; and
the second housing further includes a guide rail disposed at the second interface and in communication connection to the computing device,
when the first housing is connected to the second housing, the one or more contact points are in contact with the guide rail such that the microphone array is in communication connection to the computing device.