Voice Stress Analysis

Abstract—Speech in humans is generated by a series of formants

generated by the resonances in the nasal, pharyngeal and vocal

tracts. Linear prediction of speech is one of the most widely used

techniques for analysis of speech samples. Voice stress analysis as

a field of study holds immense importance to human life. Stress

analysis has been used in critical situation endangering life and in

several medical scenarios as well. This paper uses Linear

Predictive Coding (LPC) to analyse the stress levels in the voice

samples by comparing them to existing samples and outputting a

human emotion.

Keywords—LPC, MATLAB, stress analysis, emotion

recognition.

Linear Predictive Coding (LPC) was chiefly created for

audio and speech

 LPC analysis: It converts each frame into LPC

parameter set by using Durbin’s method[8].

Fig. 2. Different channels in the communication of emotion

B. Spectral feature extraction using LPC

Dynamic Source Routing is similar to AODV because it

forms a route only when needed or on-demand, however,

unlike AODV; DSR relies on source routing rather than table

driven routing. Source routing collects the data of every node

during route discovery and stores it in the source node. This

learned information is used by nodes to create a cached path to

the destination. This creates a redundancy wherein a large

number of routing tables are created. This problem was eli.

Fig. 3. Steps involved in LPC

LPC is a tool used for representing spectral envelope of

speech signal in compressed form. It is mostly used in audio

signal processing and speech processing. LPC gives an

accurate estimate of speech parameter. It approximates speech

samples as a linear combination of past speech samples.The

predictor coefficients, a unique set of parameters are

determined through minimizing the amount of squared

differences between the actual parameters and predicted

ones[2].LPC also helps in saving bits.

5. Speech signal, its energy contour and pitch contours

III. FEATURE SELECTION

To identify the features which contribute more to in the

classification out of the derived features which are input to

classifiers.This determines those properties of speech or

features are helpful in discriminating the emotions and

improves classification accuracy[4]. The process of searching

for subset of features that give best classification is time

consuming .So forward selection and backward elimination is

used to find the features that provide best classification.

Forward selection sequentially add one feature at a time that

most increases or least decreases classification accuracy.

Backward features starts with all features and sequentially

deletes the next feature that most decreases or least increases

classification accuracy[6].

IV. SIMULATION RESULTS IN MATLAB

MATLAB R2008b is used for feature extraction and

classification.Features are extracted from statistical moments

of the sequence. To select informative features forward feature

selection algorithm is used.

Accuracy is calculated as the ratio of correctly labeled

samples to the total