System and method for anomaly detection and extraction

Patent metadata

Field	Value
Publication number	US-9786275-B2
Application number	US-201314385670-A
Country	US
Kind code	B2
Filing date	Mar 15, 2013
Priority date	Mar 16, 2012
Publication date	Oct 10, 2017
Grant date	Oct 10, 2017

Abstract

The present invention relates to a system for suppressing transient interference from a signal. The system includes a modeling system, wherein the modeling system constructs a model of transient interference from a first signal, and a filtering system, wherein the filtering system suppresses transient interference from a second signal by applying the model to the second signal.

First claim

What is claimed:

1. A non-transitory computer-readable medium with instructions stored thereon, that when executed by a processor, suppresses transient interference from a signal, by performing the steps, comprising:
obtaining a training recording of at least one transient;
computing a training set { λ t ( l )} l = 1 M of M spectral variance feature vectors;
obtaining a recording of a signal and computing a set {λ y (l)} l=1 M of M spectral variance feature vectors;
building at least one local model for each transient type;
modeling a structure of the at least one local model as a graph;
defining a filter from the graph; and
suppressing the transient from the signal by applying the filter;
wherein the at least one local model is built according to

P
i

⁡

(

λ
y

⁡

(
l
)

)

=

η
_

i

+

∑

j
=
1

L

⁢

〈

log
⁡

(

λ
y

⁡

(
l
)

)

-

η
_

i

,

v
_

i
,
j

〉

⁢

v
_

i
,
j

;

(
13
)

wherein λ y (l) represents a vector of spectral variance values of the measured signal corresponding to timeframe l;
wherein η i represents a constant vector;
wherein υ i,j represents a set of principal eigenvectors; and
wherein the signal is a signal type selected from a group consisting of audio, speech, diagnostics for machinery, and diagnostics based on medical time series or images.

2. The non-transitory computer-readable medium of claim 1 , wherein the stored instructions further comprise the step of defining local filters from the at least one local model, wherein the local filters are defined according to
d i (λ y ( l ), λ y ( l ′))=∥ P i (λ y ( l ))− P i (λ y ( l ′))∥  (14);

wherein P i represents a linear projection of each spectral feature vector onto the local model of the ith transient type.

3. The non-transitory computer-readable medium of claim 2 , wherein the stored instructions further comprise the step of computing a non-symmetric kernel matrix.

4. The non-transitory computer-readable medium of claim 3 , wherein the non-symmetric kernel matrix is W of size M× M according to

W

l
,

l
_

=

exp
⁢

{

-



log
⁡

(

λ
y

⁡

(
l
)

)

-

log
⁡

Assignees

• Univ Yale

Inventors

• Coifman Ronald R
• Haddad Ali
• Talmon Ronen

Classifications

• G10L19/025

  Detection of transients or attacks for time/frequency resolution switching · CPC title

• G10L21/0208 Primary

  Noise filtering · CPC title

• H04R2225/43

  Signal processing in hearing aids to enhance the speech intelligibility · CPC title

• G10L15/20 Primary

  Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech (G10L21/02 takes precedence) · CPC title

• G10L15/063

  Training · CPC title