公开(公告)号：US10149047B2

公开(公告)日：2018-12-04

申请号：US14308541

申请日：2014-06-18

Applicant: Cirrus Logic Inc.

Inventor： Fredrick D. Geiger ,  Bryant V. Bunderson ,  Carl Grundstrom ,  William Erik Sherwood

IPC: H04R3/00 ,  G10L21/0208 ,  G10L21/0232 ,  G10L21/0216 ,  G10L15/20 ,  G10L25/78 ,  G10L25/84

Abstract: Techniques for processing audio signals include removing noise from the audio signals or otherwise clarifying the audio signals prior to outputting the audio signals. The disclosed techniques may employ minimum mean squared error (MMSE) analyses on audio signals received from a primary microphone and at least one reference microphone, and to techniques in which the MMSE analyses are used to reduce or eliminate noise from audio signals received by the primary microphone. Optionally, confidence intervals may be assigned to different frequency bands of an audio signal, with each confidence interval corresponding to a likelihood that its respective frequency band includes targeted audio, and each confidence interval representing a contribution of its respective frequency band in a reconstructed audio signal from which noise has been removed.

公开(公告)号：US20180240472A1

公开(公告)日：2018-08-23

申请号：US15960140

申请日：2018-04-23

Applicant: Cirrus Logic, Inc.

Inventor： Earl Vickers ,  Fredrick D. Geiger ,  Erik Sherwood

IPC: G10L21/0264 ,  G10L25/60 ,  G10L21/0224 ,  G10L25/84 ,  G10L25/78 ,  G10L25/30 ,  G10L15/06

CPC classification number: G10L21/0264 ,  G10L21/0224 ,  G10L25/30 ,  G10L25/60 ,  G10L25/78 ,  G10L25/84 ,  G10L2015/0636

Abstract: A “running range normalization” method includes computing running estimates of the range of values of features useful for voice activity detection (VAD) and normalizing the features by mapping them to a desired range. Running range normalization includes computation of running estimates of the minimum and maximum values of VAD features and normalizing the feature values by mapping the original range to a desired range. Smoothing coefficients are optionally selected to directionally bias a rate of change of at least one of the running estimates of the minimum and maximum values. The normalized VAD feature parameters are used to train a machine learning algorithm to detect voice activity and to use the trained machine learning algorithm to isolate or enhance the speech component of the audio data.