SYSTEM AND METHOD FOR LOW-LOSS REMOVAL OF STATIONARY AND NON-STATIONARY SHORT-TIME INTERFERENCES

摘要

According to a preferred aspect of the instant invention, there is provided a system and method that allows the user to remove stationary and not stationary short time interference signals in speech or other recordings in the audio part of a video recording. The system is automatically analyzing the input signal and in a plurality of individual steps detects the interference signal and removes it accordingly.

主权项

1. A method of attenuating a noise signal within an input signal containing a target signal, and said noise signal, comprising the steps of:
a. obtaining a digital representation of the input signal; b. obtaining a digital representation of the noise signal; c. partitioning said digital representation of the input signal into a plurality of overlapping input windows; d. selecting one or more of said input windows; e. calculating a input frequency spectrum from each of said selected input windows; f. determining two or more dominant peaks within each of said input window spectra, each of said dominant input frequency peaks corresponding to a dominant signal frequency; g. partitioning said digital representation of the noise signal into a plurality of overlapping noise windows; h. selecting one or more of said noise windows; i. calculating a noise frequency spectrum from each of said selected noise windows; j. using any of said calculated noise frequency spectra to identify two or more dominant noise frequency peaks and a corresponding two or more dominant noise frequencies; k. for each of said input windows, matching its dominant signal frequency peaks with said dominant noise frequency peaks; l. for each of said input windows, obtaining a count of a number of dominant signal frequency peaks that match said dominant noise frequency peaks; m. calculating an average count from all of said counts obtained from each of said input windows; n. identifying each input window whose count is greater than said average count; o. for each identified input window, using at least a portion of said matching frequencies within said identified input window to construct a transfer function for said identified input window and applying said transfer function to said identified input window, thereby producing a filtered block; p. combining each of said filtered blocks corresponding to each of said input windows and any of said input windows not identified to form a filtered input signal, thereby attenuating said noise signal within said target signal relative to said target signal; and, q. performing at least a portion of said filtered input signal for a user.