Dynamic relative transfer function estimation using structured sparse Bayesian learning
The use of a dynamic Relative Transfer Function (RTF) between two or more microphones may be used to improve multi-microphone speech processing applications. The dynamic RTF may improve speech intelligibility and speech quality in the presence of environmental changes, such as variations in head or body movements, variations in hearing device characteristics or wearing positions, or variations in room or environment acoustics. The use of an efficient and fast dynamic RTF estimation algorithm using short burst of noisy, reverberant mic recordings, which will be robust to head movements may provide more accurate RTFs which may lead to a significant performance increase.
1. A hearing device for processing signals, the system comprising:
a first transducer to transduce a first audio source into a first signal;
a second transducer to transduce a first audio source into a second signal; and
a processor configured to execute instructions to:
determine an estimated Relative Transfer Function (RTF) based on the first signal and the second signal using a hierarchical Bayesian framework;
determine a target signal based on the estimated RTF; and
generate a noise reference signal based on the first signal, the second signal, and a cancellation of the target signal.
2. The hearing device of claim 1 , wherein the hearing device includes a hearing assistance device.
3. The hearing device of claim 1 , wherein the hierarchical Bayesian framework includes a unified treatment of sparse early reflection and an exponential decaying reverberation in a prior distribution.
4. The hearing device of claim 1 , wherein the processor is further configured to execute instructions to:
iteratively determine a Relative Impulse Response (ReIR) point estimate until the ReIR point estimate converges; and
determine, in response to ReIR point estimate converging, the estimated RTF based on the ReIR.
5. The hearing device of claim 4 , wherein the processor is further configured to execute instructions to update a plurality of prior Bayesian distribution parameters based on application of Expectation-Maximization (EM) to the reverberation tail and the estimated RTF.
6. The hearing device of claim 1 , wherein:
the first signal includes a first dataset of a first duration;
the second signal includes a second dataset of a second duration; and
the first duration is substantially similar to the second duration.
7. The hearing device of claim 6 , wherein the first duration is less than 200 milliseconds and greater than 100 milliseconds.
8. The hearing device of claim 1 , further including a communication device to receive a voice activity detection input based on a Voice Activity Detector (VAD), wherein determining the estimated RTF is further based on the voice activity detection input.
9. The hearing device of claim 1 , wherein determining a noise reference signal based on the cancellation of the target signal includes cancelling the target signal based a blocking matrix of an adaptive Generalized Sidelobe Canceler, the blocking matrix designed using the RTF.
10. A method for processing signals, the method comprising:
receiving a first signal from a first transducer of a hearing device;
receiving a second signal from a second transducer;
determining an estimated Relative Transfer Function (RTF) based upon the first signal and the second signal using a hierarchical Bayesian framework;
determining a target signal based on the estimated RTF;
determining a noise reference signal based on the first signal, the second signal, and a cancellation of the target signal; and
cancelling interference based on the noise reference signal.
11. The method of claim 10 , wherein the hearing device includes a hearing assistance device.
12. The method of claim 10 , wherein a unified treatment of sparse early reflection and an exponential decaying reverberation in a prior distribution is incorporated into the hierarchical Bayesian framework.
13. The method of claim 10 , wherein determining the estimated RTF includes:
iteratively determining a Relative Impulse Response (ReIR) point estimate until the ReIR point estimate converges; and
determining, in response to ReIR point estimate converging, the estimated RTF based on the ReIR.
14. The method of claim 13 , wherein iteratively determining the ReIR point estimate includes interactively updating a plurality of prior Bayesian distribution parameters based on application of Expectation-Maximization (EM) to the reverberation tail and the estimated RTF.
15. The method of claim 10 , wherein:
the first signal includes a first dataset of a first duration;
the second signal includes a second dataset of a second duration; and
the first duration is substantially similar to the second duration.
16. The method of claim 15 , wherein the first duration is less than 200 milliseconds and greater than 100 milliseconds.
17. The method of claim 10 , wherein determining the estimated RTF is performed by a processor within the hearing assistance device.
18. The method of claim 10 , wherein determining the estimated RTF is performed by a processor within a computing device wirelessly connected to the hearing assistance device.
19. The method of claim 18 , further including:
generating a voice activity detection input based on a Voice Activity Detector (VAD); and
wherein determining the estimated RTF is further based on the voice activity detection input.
20. The method of claim 10 , wherein determining a noise reference signal based on the cancellation of the target signal includes cancelling the target signal based a blocking matrix of an adaptive Generalized Sidelobe Canceler, the blocking matrix designed using the RTF.