Вы находитесь на странице: 1из 4

Project Proposal

SPEECH ENHANCEMENT USING STASTICAL


MODELLING
SECTION 1: SPECIFICATION DATE

July 2009

SECTION 2: CLIENT

GALAXIE SOFTWARE SOLUTIONS

SECTION 3: PROJECT TYPE

APPLICATION DEVELOPMENT

SECTION 4 : DESCRIPTION

Speech enhancement deals with the reconstruction of original speech


signal from a corrupted, noisy version. In general, there exists a need for
digital voice communications, human-machine interfaces, and automatic
speech recognition systems to perform reliably in noisy environments. For
example, in hands-free operation of cellular phones in vehicles, the speech
signal to be transmitted may be contaminated by noise. In many cases, these
systems work well in nearly noise-free conditions, but their performance
deteriorates rapidly in noisy conditions. The noise encountered in such
situations can be non-stationary and potentially non-Gaussian, also speech is
generally modeled as Gaussian, whereas in reality a super-Gaussian
distribution like Laplacian or Gamma would be a much better fit. Classical
work on speech enhancement has revolved around removing additive
Gaussian noise, thus the need for using statistical techniques to solve this
problem.
The different techniques used for speech enhancement include:

 Wiener Filter
 Spectral Subtraction
 Statistical Modelling

Existing System:

The Existing system for speech enhancement was winner filtering and
spectral substraction.in this case the enhancement is not up to the mark .

Proposed system:

The suitable technique for speech enhancement was introduced for low noise
extraction is Statistical modelling. In this method we are using DFT and
IDFT.

The noisy speech signal that is measured, is first transformed to the


frequency domain using DFT. The noise and clean speech spectral
coefficients’ statistics are estimated using a noise estimation technique and
clean speech is estimated using MAP estimates by assuming speech to have
a Laplacian / Gamma prior and noise to have a Gaussian / Gamma prior.
The clean speech spectral coefficients are then transformed into the time
domain using IDFT.

In this paper, we propose a statistical model for speech enhancement that


takes into account the time-correlation between successive speech spectral
components. It retains the simplicity associated with the Gaussian statistical
model, and enables the extension of existing algorithms to noncausal
estimation. The sequence of speech spectral variances is a random process,
which is generally correlated with the sequence of speech spectral
magnitudes.

In the proposed algorithm these are the a priuri SNR of each spectral
component, and the variance of each noise spectral component. It is
demonstrated here that by using different estimators for the a priori SNR,
different STSA estimations result. For example, using the “power spectral
subtraction” method for estimating the a priuri SNR results in an STSA
estimator which is nearly equivalent to the “spectral subtraction” STSA
estimator.

We show that a special case of the causal estimator degenerates to a


“decision-directed” estimator with a time-varying frequency-dependent
weighting factor. Experimental results demonstrate the improved
performance of the proposed algorithms.

SECTION 5: TECHNOLOGY

Software requirement : MATLAB 6p5


Operating system : Windows

Вам также может понравиться