Sie sind auf Seite 1von 15

Speech Enhancement

based on Spectral
Subtraction method using
Spectral Flatness Measure
Project Members:
Padmanabh Maski (2BA09EC046)
Adil e shahabaaz Ujani (2BA09EC001)
Deepak Naik (2BA09EC044)
PROJECT GUIDE

PROF. SHRIDHAR .K
IDDALAGI.

HOD

PROF. S.M.

Date-

Introduction to Speech:
Dynamic

information bearing signal,


also called Acoustic waveform.
Most desirable medium of
communication between human
beings.
Characteristics:

Band limited (Band width=4KHz)


Non zero auto correlation.
Non flat spectra.
Fundamental frequency range 80Hz350Hz.

Speech Processing:
Study

of speech signals and


processing methods of these
signals.
Aspects: Acquisition, Manipulation,
storage, Transfer, Output of digital
speech signal.
Applications:
Recognition
Synthesis
Compression of human speech.

Speech Enhancement:
Method

of suppression of noise.
Improve intelligibility or overall
perceptual quality of degraded
speech signal.
Application:

mobile communication
teleconferencing systems
speech recognition
hearing aids
Speech coding, etc.

Types of enhancement
systems:

Suppression of noise using


periodicity of speech.
Model based speech enhancement.
Short term spectral amplitude
technique(STSA).
Spectral

subtraction:

Method of noise reduction based on


STSA.
Simplest algorithm to eliminate the
background stationary noise.

Spectral Flatness
Measure:

Measure of noisiness or sinusoidility


of spectrum.
Describes the flatness properties of
the spectrum of an audio signal.
Lies in the range 0 to 1.
For tonal signal close to 0 and for
noisy signal close to 1.
Ratio of geometric mean to
arithmetic mean of magnitude
spectrum.

Literature Survey:
1. Author:- Shlomo Dubnov
Title:- Generalization of Spectral
Flatness Measure for NonGaussian Linear Processes.

Generalization of the standard


spectral flatness measure using
an information theoretic
formulation of randomness.
Applied to improve voicing
determination in speech signals.

2. Author:- Jiri Pribil and Anna


Pribilova
Title:- Spectral Flatness Analysis for
Emotional Speech Synthesis and
Transformation.

Aimed at statistical analysis of


spectral flatness in three emotions
(joy, sadness, anger).
SFM used to identify different
emotions.

3. Author:- Alexander Friedlander.


Title:- Content based identification
of audio material using MPEG-7 low
level description.

Presents system for reliable, fast


and robust identification of audio
material.
SFM used as discriminating criterion
between different audio signals.

4. Authors:- Robert Yantorno, Kasturi


Rangan, Jereme M.
Title:- The spectral autocorrelation
peak valley ratio - A usable speech
measure employed as a co-channel detection
system.

An

effective method to identify the


existence of co-channel speech.
Voiced portions identified using
spectral flatness measure.

5. Authors:- Christian Uhle, Thomas


Sporer.
Title:- Extraction of drum tracks from
polyphonic music using independent
subspace analysis.

Separated tracks have sufficient


audio quality.
The amount of harmonic sustained
sound is tolerable.
SFM describes the flatness properties
of the spectrum of an audio signal.

Outcome of Literature
Survey

SFM provides better discrimination


between voiced and unvoiced speech.
SFM expresses the deviation of signals
power spectrum over frequency from
a flat shape.
One can vary the threshold for voiced
speech using spectral flatness.
No need to consider changing the
zero-crossing threshold.

Problem formation:

Use of SFM approach to


overcome the drawbacks
present in Spectral subtraction
method for speech
enhancement.

Reference:
Generalization

of Spectral Flatness Measure


for Non-Gaussian Linear Processes by
Shlomo Dubnov.
Spectral Flatness Analysis for Emotional
Speech Synthesis and Transformation by Jiri
Pribil and Anna Pribilova.
Content based identification of audio
material using MPEG-7 low level description
by Alexander Friedlander.
The spectral autocorrelation peak valley
ratio - A usable speech measure employed
as a co-channel detection system by Robert
Yantorno, Kasturi
Rangan,
Jereme M.
Extraction of drum tracks from polyphonic
music using independent subspace analysis
by Christian Uhle, Thomas Sporer .

THANK YOU

Das könnte Ihnen auch gefallen