1

SPEECH RECOGNITION

Seminar By: Suraj Vitthal Gaikwad

07-Feb- Guided By: Prof. S. R. Lahane
2013
Outline
2

 Introduction
 Speech Recognition Process
 Types Of Speech Recognition Systems
 Algorithms
 Applications
 Advantages & Disadvantages
 Future Scope
 Conclusion

SPEECH RECOGNITION 07-Feb-13

Introduction
3

 Speech recognition is the process by

which a computer (or any other type of
machine) identifies spoken words.
 Basically, it means talking to your
computer, AND having it correctly
understand what you are saying.
 An alternative to traditional methods of
interacting with a computer.

SPEECH RECOGNITION 07-Feb-13

4

SPEECH RECOGNITION 07-Feb-13

Speech Recognition Process
5
 Signal Processing
 Convert the audio wave into a sequence of feature
vectors
 Speech Recognition
 Decode the sequence of feature vectors into a
sequence of words
 Semantic Interpretation
 Determine the meaning of the recognized words
 Dialog Management
 Correct the errors and help get the task done
 Response Generation
 What words to use so as to maximize user
understanding
 Speech Synthesis (Text to Speech)
SPEECH
 RECOGNITION
Generate synthetic speech from a 07-Feb-13
‘marked-up’ word
Typical Speech Recognition
6
Process

SPEECH RECOGNITION 07-Feb-13

Types of Speech
7
Recognition
 Isolated Words
 Single utterance at a time
 Connected Words
 Separate utterances together with a
minimal pause between them
 Continuous Speech
 Rehearsed speech or dictation
 Spontaneous Speech
 Natural speech
SPEECH RECOGNITION 07-Feb-13
Algorithms
8

 Dynamic Time Warping

 an algorithm for measuring similarity
between two sequences which may vary
in time or speed.
 Hidden Markov Models
 Neural Networks

SPEECH RECOGNITION 07-Feb-13

Hidden Markov Model
9

 In a HMM, the state is not directly visible,

but output, dependent on the state, is
visible.
 Each state has a probability distribution
over the possible output tokens.
Therefore the sequence of tokens
generated by an HMM gives some
information about the sequence of
x — states
states. y — possible observations
a — state transition
probabilities
b — output probabilities
SPEECH RECOGNITION 07-Feb-13
HMM Example
10

SPEECH RECOGNITION 07-Feb-13

Neural Network
11

 A neural network consists of an interconnected

group of artificial neurons, and it processes
information using a connectionist approach
to computation.
 An NN is typically defined by three types of
parameters:
 The interconnection pattern between different
layers of neurons
 The learning process for updating the weights of the

interconnections
 The activation function that converts a neuron's

weighted input to its output activation.

SPEECH RECOGNITION 07-Feb-13
Speech Recognition
12
Softwares
 Open source
 Julius
 Macintosh
 Dragon Dictate
 Mobile Devices/ Smartphone
 Google Now
 Siri
 Micromax AISHA
(Artificial Intelligence Speech Handset Assistant)
 S Voice
 Iris (Intelligent Rival Imitator of Siri)
 Windows
 Dragon NaturallySpeaking
 Windows Speech Recognition
SPEECH RECOGNITION 07-Feb-13
Applications
13

 Games and Edutainment

 Data Entry
 Document Editing
 Speaker Identification/Verification
 Automation at Call Centers
 Medical/Disabilities
 Fighter Aircrafts

SPEECH RECOGNITION 07-Feb-13

Advantages
14

 Increases Productivity
 Can help with menial computer tasks
 Can help people with disabilities
 Cost Effective
 Diminishes Spelling Mistakes

SPEECH RECOGNITION 07-Feb-13

Disadvantages
15

 Inaccuracy & Slowness

 Vocal Strain
 Adaptability
 Out-of-Vocabulary (OOV) Words
 Spontaneous Speech. Etc
 Accent, Dialect and Mixed Language

SPEECH RECOGNITION 07-Feb-13

Future Scope
16

 Achieving efficient speaker independent

word recognition
 SRS may have the ability to distinguish
nuances of speech and meanings of
words.
 Stand alone Speech Recognition
Systems.
 Wearable Speech Recognition System.
 Talk with all the devices.
SPEECH RECOGNITION 07-Feb-13
Conclusion
17

 Within five years, speech recognition

technology will become so pervasive in
our daily lives that service environments
lacking this technology will be
considered inferior.
 Speech recognition will revolutionize the
way people interacted with Smart
devices & will, ultimately, differentiate
the upcoming technologies.

SPEECH RECOGNITION 07-Feb-13

References
18
 JOE TEBELSKIS {1995}, SPEECH RECOGNITION USING NEURAL
NETWORKS, School of Computer Science, Carnegie Mellon University
 KÅRE SJÖLANDER {2003}, An HMM-based system for automatic
segmentation and alignment of speech, Umeå University, Department
of Philosophy and Linguistics
 KLAUS RIES {1999}, HMM AND NEURAL NETWORK BASED SPEECH
ACT DETECTION, International Conference on Acoustics and Signal
Processing (ICASSP’99)
 B. PLANNERER {2005}, AN INTRODUCTION TO SPEECH RECOGNITION
 KIMBERLEE A. KEMBLE, AN INTRODUCTION TO SPEECH RECOGNITION,
Voice Systems Middleware Education, IBM
 LAURA SCHINDLER {2005}, A SPEECH RECOGNITION AND SYNTHESIS
TOOL, Department of Mathematics and Computer Science, College of
Arts and Science, Stetson University
 MIKAEL NILSSON, MARCUS EGNARSSON {2002}, SPEECH
RECOGNITION USING HMM, Blekinge Institute Of technology
SPEECH RECOGNITION 07-Feb-13
19

THANK
YOU…!!
ANY QUESTIONS…??

SPEECH RECOGNITION 07-Feb-13