Browse Prior Art Database

%BLT% Restoration of clipped signals with application to speech recognition

IP.com Disclosure Number: IPCOM000240971D
Publication Date: 2015-Mar-16
Document File: 1 page(s) / 47K

Publishing Venue

The IP.com Prior Art Database

Abstract

This article has not been copy edited and is not ready for publication. Please see note below.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 1

%

%BLT

BLT%

Restoration of clipped signals with application to speech recognition

      Restoration of clipped signals with application to speech recognition
We consider the problem of signal clipping and we study the performance of speech recognition in the presence of clipping. Assuming that the undistorted signal is band-limited, we develop an iterative algorithm for restoring the corrupted samples, whose MSE is monotonically non increasing. In a different approach, we model the signal as an auto-regressive process and as a by product of the estimation-maximization algorithm for estimating the model parameters, the signal is recovered. The effects of these methods on the accuracy of speech recognition is studied.

Clipping is a form of non-linear distortion that produces a flat cut-off once the signal exceeds a threshold. In the frequency domain, it produces strong harmonics in the high frequency range. There are several causes of clipping. When an amplifier is pushed to create a signal with more power than its power supply can produce, it will amplify the signal only up to its maximum capacity , simply cutting off the extra signal beyond the capability of the amplifier. In digital signal processing, clipping occurs when the signal is restricted by the range of a chosen representation.

Clipping distorts the signal in a way that cannot be perfectly restored . The mapping is many to one and the information contained in the peaks that are clipped is completely eliminated , providing only a range of possible values that the signal may have taken. It is therefore preferable to avoid clipping in the first place . Lightly clipped bandwidth limited signals that are highly oversampled have a high likelihood of perfect repair.

An algorithm for the restoration of clipped speech signal is proposed in [1], where the clipped speech is assumed to be linearly predicted with high accuracy. The prediction coefficients are computed either by solving a least-squares problem or b...