Noise filtering device
AssigneeRockwell International Corporation
InventorsJames E. Paul, Visvaldis A. Vitols
Life of this patent
Estimating the term…
Abstract
Noise such as hum (whose components are self-correlated over a long time-interval) is estimated (separated) from Signal such as speech (whose components are self-correlated only over a short time-interval) by using a time-domain transversal filter, whose tap-weight coefficients may be fixed (Wiener filter) or variable (Adaptive filter). A combiner combines the estimated noise and original input signal to develop an output speech signal free of noise. The output speech may by further applied to a spectrum shaper to restore a natural shape.
Claims12
A device for removing undesired correlated components from an input degraded audio signal, said device comprising:
- filter means responsive to the input audio signal for substantially estimating only the undesired correlated components of the input audio signal;
- and
- a combiner for substracting the input audio signal from the estimate of the undesired correlated components to develop an output audio signal containing substantially no undesired correlated components.
A filter device comprising:
- means for developing a least mean square estimate of the correlated components of an input signal which contains both correlated and uncorrelated signal components;
- and
- means for combining the least mean square estimate with the input signal to develop an output signal containing only uncorrelated signal components.
A machine method for removing undesired correlated components from an input audio signal comprising the steps of:
- developing a least means square estimate of the undesired correlated components in the input audio signal;
- and
- subtracting the least mean square estimate from the input audio signal to develop an output audio signal containing substantially no undesired correlated components.
A machine method for removing undesired correlated components from an input degraded voice signal comprising the steps of:
- producing an estimate of the undesired correlated components;
- and
- combining the estimate of the undesired correlated components with the degraded voice signal to develop an output voice signal with the undesired correlated components removed.
Dependencies are inferred from the claim text.