Introduction - Robust Emotion Recognition Using Spectral and Prosodic Features - page 2

able to process the speech for capturing the above information sources present in
speech. While interacting with a machine, humans expect the machine to perform
above speech tasks.
Humans extensively use emotions to express their intentions through speech [ 1 ].
The same textual message would be conveyed with different semantics (meaning)
by incorporating appropriate emotions. Humans understand the intended message by
perceiving the underlying emotions in addition to phonetic information. Therefore,
there is a need to develop speech systems that can process emotions along with the
message [ 2 ]. The basic goals of emotional speech processing are (a) recognizing
the emotions present in speech and (b) synthesizing the desired emotions in speech
according to the intended message. From the machine's perspective, understanding
speech emotions can be viewed as the classification or discrimination of emotions.
Synthesis of emotions can be viewed as incorporating the emotion-specific knowl-
edge during speech synthesis. The emotion-specific knowledge is acquired from the
emotion models, designed for capturing the emotion-specific characteristics.
In this topic, we are focussing on characterization and discrimination of emotions
by a machine using speech signal. Emotion processing is very complex task. It
involves semantics at higher level, intention of themessage and natural characteristics
of a speaker. Expression of emotions depends on several factors such as speaker,
gender, language, and social constraints. Most of the existing emotion processing
systems explored the full-blown simulated emotions. But, in real life emotions are
not always full-blown, and mostly blended emotions. It is very difficult to crisply
discriminate the natural emotions. Most of the works have explored the conventional
spectral and prosodic features for characterizing the emotions. The extreme emotions
such as anger and sadness may be discriminated well by using the conventional
spectral and prosodic features, but they may not be effective for discriminating other
emotions. For affective discrimination of emotions, in this topic we have proposed
some new spectral and prosodic features. The proposed features are expected to
capture the finer variations in emotion specific spectral and prosodic characteristics.
The proposed features are evaluated for discriminating the real life emotions.
1.2 Emotion from Psychological Perspective
Emotions have been studied in several scientific disciplines, such as: Biology (Phys-
iological), Psychology, Speech science, Neuroscience, Psychiatry, Anthropology,
Sociology, Communication and so on. Subjects like Business Management and
Advertising need extensive use of emotion processing. As a result, distinctive per-
spectives on the concept of emotion have emerged, appropriate to the complexity and
variety of the emotions. However, it is important to consider these different perspec-
tives not as competitive but as complementary. In this topic, emotions are analyzed
in view of psychological, physiological and speech perspectives.
Psychology of emotions can be viewed as a complex experience of conscious-
ness (psychology), bodily sensation (physiology), and behavior (action-speech).