New

24-09-2014 CSE DEPARTMENT VAST 1
PRESENTED BY :
DEVIKA M D
ROLL NO. : 6
MTECH CSE(14-16)
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING
Vidya Academy of Science and Technology
Thalakkottukara, Thrissur – 680 501
Sentence-Based Sentiment Analysis
for Expressive Text-to-Speech

CONTENTS
 Introduction
 Sentiment Analysis for TTS system
 Expressive TTS system
 Text to Speech System
 Relevance
 Future scope
 Conclusion
 References

INTRODUCTION
 Natural language processing
 Sentiment Analysis for TTS system
 Text to Speech system
 Expressive TTS system

Sentiment Analysis For TTS Purposes
WHAT IS SENTIMENT/OPINION?
Sentiment
 Feelings
 Attitudes
 Emotions
 Opinions
An opinion is a personal belief or judgment that is not
founded on proof or certainty .

WHAT IS SENTIMENT ANALYSIS?
 Computational study of opinions, sentiments,
 Evaluations, attitudes, appraisal, affects, views,
emotions, subjectivity, etc., expressed in text.
 Identify the orientation of opinion in a piece of
text

 Two main types of textual information.
- Facts and Opinions
Note: Factual statements can imply opinions too.
 Mainly because of the Web
 Huge volumes of opinionated text.
Contd...

Two types of opinion
 Direct sentiment expressions on some target objects,
E.g., products, events, topics, persons. E.g., “the
picture quality of this camera is great.”
 Comparative Opinions: Comparisons expressing
similarities or differences of more than one object.
Usually stating an ordering or preference.
E.g. “car x is cheaper than car y.”
Contd...

Text
classifier
Text processing
Feature
extraction
Positive info Neutral info Negative info
Contd...

TEXT PROCESSING
tokenization
Removing
stop words
Symbol
analysis

Tokenization
 involves splitting the text by spaces, forming a list of
individual words per text
 called a bag of words
Removing stop words
 Remove stopwords from bag of words
 E.g. : also, etc. , able ,or ,and
Symbol analysis
 E.g. :- smileys can indicate emotion
Contd...

 Feature extraction
unigram
Bigram/trigr
am
Neutral
tweets

 A unigram is simply an N-gram of size one, or a single
word.
 Bigrams and trigrams from our tweets as features to
train our classier.
Eg: don’t like ,not happy
 Neutral tweets – tweets that doesn’t have any
particular sentiment.
 Lexicon - which is a list of words that are predefined
with a sentiment, either positive or negative
Contd...

LEVELS
Document level
Sentence level
Word level

How Sentiment Analysis?
 Emolib- extract the affect from text according to the
feelings written in text.
 System is designed with a pipeline.
Text
Emolib
pipeline
Tag
Eg; I hate
you
Negative
sentence

Emolib pipeline

Lexical analyser
 covert plain text to tokens.
 filter out “stop words”.
 produced with javcc2.
Sentence splitter
 sentence to binary tree.
 examine uppercase letters, exclamation, question
marks etc.
Contd...

POS Tagger
 determine nouns, verb and adjectives.
 implemented using Standord log linear.
Word Sense Disambiguator
 determines correct sense of a word according to the
context.
 implemented using word net ontology.
Contd...

Stemmer
 group those word share a common meaning.
 use Porter stemming algorithm.
 Keyword Spotter
 emotional dimensions to emotional word
 use ANEW corpus.
Contd...

Average Calculator
 calculate average emotional dimensions.
 AM of dimension at sentence level.
Classifier
 Labels the text with appropriate emotion.
 predicts appropriate sentiment label to the text.
Formatter
 present result in usable form(XML )
Contd...

Datasets
Determine effective emolib configuration
Semeval 2007 dataset
Twitter dataset

Contd...

Sentiment Analysis Tools
 Emoticons.
 LIWC -Linguistic Inquiry and Word Count.
 SentiStrength.
 SentiWordNet.
 SASA -SailAil Sentiment Analyzer .
 Happiness Index.
 PANAS-t.

WHY SENTIMENT ANALYSIS?
 Movie: is the review positive or negative?
 Products : what do people think about a new phone?
 Survey : how is consumers responding to a product ?
 Politics : what do people think of a political issue?
 Prediction : predict election outcomes /market trends

Expressive TTS system
• Deliver expressive cues when synthesizing.

Text To Speech System
Text To Speech translation
 Automatic production of speech.

 NLP: Capable of producing a phonetic transcription of the
first read, together with desired intonation and rhythm.
 DSP: Transforms the symbolic information it receives in
to speech.
Contd...

The NLP component

The Text analysis
Pre-processing module
 input sentences into lists of words.
 identifies numbers, abbreviations, acronyms.
Morphological analysis module
 Identifies morphens.
Contd...

Contextual analysis module
 considers words in their context.
Syntactic-prosodic parse
 examines the remaining search space and finds the
text structure.
Contd...

Automatic phonetization
The Letter-To-Sound (LTS) module- automatic
determination of the phonetic transcription of the
incoming text.
 Dictionary based system
 Rule based system transfer
Contd...

Prosody generation
 properties of the speech signal which are related to
audible changes in pitch, loudness.
Contd...

The DSP component
 Rule-based synthesizers

 Concatenative synthesizers
Contd...

Relevance
 Sentiment analysis is the process of identifying
people’s attitudes and emotional states from language.
 Determine review on upcoming movie, correlating
statements about a political party with people’s
likeliness to vote for that party, restaurant reviews

Future work
 Need to increase the size data.
 Need to analyse shortforms and alternative sentences.

CONCLUSION
 Sentiment analysis can be successfully used to convert
large amount of unstructured data in to useful
information.
 Sentiment analysis aims to determine the attitude of a
speaker or a writer .

REFERENCE
[1] N. Campbell, “Conversational speech synthesis and the need for some laughter,” IEEE Trans. Audio, Speech,
Lang. Process., vol. 14, no. 4, pp. 1171–1178, Jul. 2006.
[2] R. Calix, S. Mallepudi, B. Chen, and G. Knapp, “Emotion recognition in text for 3-D facial expression
rendering,” IEEE Trans. Multimedia, vol. 12, no. 6, pp. 544–551, Oct. 2010.
[3] T. Wilson and G. Hofer, “Using linguistic and vocal expressiveness in social role recognition,” , pp. 419–
422, 2011.
[4] F. Alías, X. Sevillano, J. C. Socoró, and X. Gonzalvo, “Towards high-quality next-generation text-to-speech
synthesis: A multidomain approach by automatic domain classification,” IEEE Trans. Audio, Speech, Lang.
Process., vol. 16, no. 7, pp. 1340–1354, Sep. 2008.
[5]J. Bellegarda, “A data-driven affective analysis framework toward naturally expressive speech synthesis,”
IEEE Trans. Audio, Speech, Lang. Process., vol. 19, no. 5, pp. 1113–1122, Jul. 2011.

[6] J. Pitrelli, R. Bakis, E. Eide, R. Fernandez,W. Hamza, and M. Picheny, “The IBM expressive
text-to-speech synthesis system for American English,” IEEE Trans. Audio, Speech, Lang.
Process., vol. 14, no. 4, pp. 1099–1108, Jul. 2006.
[7] B. Pang and L. Lee, “opinion mining and sentiment analysis,” Found at. Trends in Inf.
Retrieval, vol. 2, no. 1–2, pp. 1–135, 2008.
[8] V. Francisco, R. Hervás, F. Peinado, and P. Gervs, “EmoTales: Creating a corpus of folk tales
with emotional annotations,” Lang. Res. Eval., vol. 45, pp. 1–41, Feb. 2011.
[9] A. R. F. Rebordao, M. A. M. Shaikh, K. Hirose, and N. Minematsu, “How to improve TTS
systems for emotional expressivity,” in Proc. Inter speech’09, Sep. 2009, pp. 524–527.
[10] C. O. Alm, D. Roth, and R. Sproat, “Emotions from text: Machine learning for text-based
emotion prediction,” in Proc. HLT’05, 2005, pp. 579–586.
Contd...

QUESTIONS

THANK YOU

New

More Related Content

Viewers also liked (13)

Similar to New (20)

Recently uploaded (20)

New