WORKSHOP ON PRONUNCIATION LEXICONS FOR INDIC LANGUAGES

notownbuffAI and Robotics

Nov 17, 2013 (3 years and 9 months ago)

60 views





WORKSHOP ON PRONUNCIATION LEXICONS

FOR INDIC LANGUAGES





Organized by W3C India in close co
-
ordination with IIT
Kharagpur and JNU

4
-
5 May 2011
,
Jawaharl
al

Nehru University



Introduction

The accurate specification of pronunciation is critical to the

success of speech applications.
Most Automatic Speech Recognition (
ASR
) and Text
-
To
-
Speech (
TTS
) engines intern
ally
provide extensive high quality

lexicons

with pronunciation information for many words or
phrases. To ensure a maximum coverage of the words or phrases used by an application,
application
-
specific pronunciations may be required.

The complexity of Indian scenario is augmented manifold, as India has 22 constitutionally
recognized languages and we need to develop solutions in all these languages.


One of the important components i
n the development of speech technology for a particular
language is the pronunciation dictionary. It represents the interface between speech
analysis and speech interpretation.

.


Most state
-
of
-
the
-
art Automatic Speec
h Recognition (ASR) and Text
-
To
-
Speech (TTS)
systems rely on pronunciation lexicons, which contain pronunciation information on words,
multi
-
word expressions and even phrases that commonly occur in a given domain of
application.


Several guidelines have b
een reported to define the structure of a pronunciation lexicon.
The W3C Voice Browser Activity has recently published a Pronunciation Lexicon
Specification (PLS) Version 1.0. PLS is designed to enable interoperable specification of
pronunciation informati
on for both speech recognition and speech synthesis engines within
voice browsing applications. The language is intended to be easy to use by developers while
supporting the accurate specification of pronunciation information for international use.









Objective

The Objective of the Workshop
includes

the following aspects of the creation of
pronunciation lexicons:



Pronunciation Lexicon Specification (PLS) of Voice XML Applications and the
Pronunciation Lexicon Markup language.



The Structure of Phonetic
Lexicons and Phonetic alphabets used in Phonetic
Lexicons.



Articulatory phonetic and Acoustic phonetic of Indian languages: An example study
of Bangla.



Orthographic and Phonic Variation: Types and Applications in TTS and ASR.



Incorporation of Phonic Variat
ion in Phonetic Lexicons with a case study of Bangla.














Agenda

Day
-
I 4
th

May

1

Welcome address:
Prof. Pramod Pandey

10.30 AM


10.35 AM

2.

Inaugural Keynote Address:
Shri N. Ravi Shanker , Joint
Secretary , Dept of Information Technology , Govt. of India

10:35 A.M.


10:45 A.M.

3.

Key Note Address:
Prof. R.N. Menon , Dean , School of
Languages , JNU , New Delhi

10 :45 A.M


10:55 A.M.

4

Technical Plenary Address by Ms Swaran Lata

10.55AM


11.30 AM


Vision on PLS Standardization in Indic Languages


Tea Break :
11:30 A.M. to 11 :45 A.M

Technical Session 1

4

Presentation By Dr. S.K. Das Mandal

11:45 A.M. to 12:30 P.M.




W3C PLS stan
dards



Comments on essential requirement/ modification and
for PLS standards with respect to Indian languages



5.

Presentation by TTS, ASR Consortium
leaders and

other
Speech project Leaders about the present Lexicon
Development,

and Data Formats.


12:30
P.M.
-
13:30 P.M.

Lunch

1.30 PM


2.15 PM

6


Presentation by Prof. Pramod Pandey



Phonological requirements in PLS specifications in Hindi
Language

2.15 PM


3.15 PM

7.

Presentation by Dr. S.K. Das Mandal



Integrated Development Environment (IDE) for
Gener
ation of Pronunciation Lexicon in Indian




languages



IPA requirements in PLS

3:15 P.M.


4:00 P.M.



Tea

4.15 PM


4.30 PM

8

Focused Deliberations on Indic Languages requirements in
PLS standard



IPA requirements of PLS



Orthographic and Phonetic Variation

of Indic
languages in PLS



Requirements inclusion of abbreviations in PLS



4.30 PM


5.30 PM

Day 2 5
th

May

1

Finalization of draft Indic Languages Re
quirements in the
form of recommendations for inclusion in PLS standard

10.30 AM


11.30 P.M.




2

Training on Integrated Development Environment for PLS
development for Indic languages


by Dr. Shyamal Das
Mandal & Shri Arup Saha



11.45 AM


1.30 A
M


Valedictory

1.30 PM


2.00 PM


LUNCH






Workshop Venue

The workshop held at

Jawaharlal Nehru University



Centre for Linguistics


New Mehrauli Road


N
ew Delhi
-
110067


Welcome Address

Prof. Pramod Pandey,

Inaugural Keynote

Address

Shri N. Ravi Shanker, Joint Secretary, Dept of Information Technology, Govt. of India

started
with
inaugural address and welcomed all the speakers as well as the delegates in
the
workshop. He talked about the
activities carried out by W3C
India in the field of
Pronunciation Lexicon for Indic Languages.

Key Note Address

Prof. R.N. Menon
, Dean, School of Languages, JNU, New Delhi

Technical Plenary Address

Ms. Swaran Lata,
Countr
y Manager,
W3C India













Testimonials

























Workshop was quite interactive and
knowledge sharing.




By S.S. Agrawal

Executive Director



Very good, having very interactive lively
sessions.



By Gopal Ram

Phd Linguistic


The programme was really informative
and useful. It was nicely arranged and
organized.


By
Leena Hunnargikar

Phd Student




This may be the 1
st

workshop in PLS for
Indic Languages. Quality is good.



By Joyanta Basu

Member technical Staff





Technical
Session Summary



Pronunciation Lexicon in W3C Framework

By
Dr
. Shyamal

Kumar Das Mandal, Professor,
IIT Kharagpur







Speech Synthesis and Pronunciation Dictionaries

By
Kishore S. Prahallad
,
Assistant Professor, IIT Hyderabad






Pronunciation Specification for Lexicon Development by ASR Consortium

By
Mr Milton Samirakshma Bepari
,
Senior Scientific Officer,
CDAC, Kolkata






He
explained

about
Pronunciation

Lexicon

Specification

(PLS)

covers how to use TTS and
ASR
b
y

the

<lexicon>

element

of

SSML
.

He was discussed about how to capture
multiple

pronunciations for the same o
rthography

under the solution of existing PLS specification

and multiple orthography
with

same pronunciations

with Bangla example
.

He was disc
ussed about Morphological Analysis for reducing the size of the Lexicon
. He
addressed the proposed solutions like as: POS as an attribute, POS as an element

for
multiple pronunciations for the same orthography.





He

addressed

the
Pronunciation

dictionar
y consists of words, Word are transcribed into a
sequence of syllables,
and Syllables

are transcribed to phones.


He discussed about the Issues in pronunciation dictionary design with the example of Hindi,
Telugu, Bengali,
and
Malayalam
.

He was also discus
sed

Pronunciation features required for
Text to Speech

Synthesis

and

Partial pronunciation dictionaries
.


He addressed the
Implement

and

deploys

a

speech

based

system

to

get

prices

of

agricultural

commodities

using

telephone/mobile
.

He discussed about
ASR

Standard Label set
,
Suffixes in
the Labels,
Additional Labels
,
to represent ‘Bindu’ in the script using nasals,
ASR Standard Tags
for Non Speech Sounds
, Extra phones used compared to ASR Standard

Labels
, Extra Non
-
Speech Sounds used
.

He addressed that t
he

experience

of

data

collection

and

transcription

for

six

languages

in

six

states,

th
e
y

feel

that

the

labels

used

by

the

consortium

are

easy

to

use

and

not

confusing.

ASR

consortium

labels

may

be

adopted

for

other

Indian

languages

with

additional

phones

whe
re

and

whenever

required.






Vision and Roadmap for PLS Standardization in Indic Languages

By
Ms. Swaran Lata, Country Manager, W3C India







Computational Phonology and Pronunciation Lexicons

By
Anil Kumar Singh
,
Assistant Professor, KIIT Bhubaneshwar







Dr. S.S Agrawal, Executive Director, KIIT GROUP OF COLLEGES





Dr. Pramod Pandey
,

JNU New Delhi

He addressed
the
various issues related to different speech sounds in Indian languages. He talked
about the phonemes and allophones and their surface boundaries. He explained about the
ORL(Orthography Relevance level) which is used to determine the phonemes and allophones.

He
explained different rules related for pronunciation by giving example of
चंदन
,
चंदा
,
कंघी
.
He also
She

discussed about the
Pronunciation

Lexicon

Specification

(PLS), Speech Interface
Framework and Indian Language Requirements.

She explained PLS definition with
their
attributes

and elements
description
, PLS structure
. She add
ressed PLS document, SSML
with PLS,
Future updations of PLS using RDF

and building a common data format for both
TTS and ASR
.

She
also
raised

the issues

for discussion such as
using IPA/SAMPA/XSAMPA
?

How to port
existing data to PLS?

Standardize metadata
,
Orthographic variation coverage
,
and
Reference

phoneme set
, etc.






























He discussed about the
Computational Phonology
,

Models of Phonology
,

Computational
Phonetic Model of

Scripts (CPMS)
,
Models of Variation
,

Letter to Phoneme Co
nversion
,
Transliteration,

Synchronic and Diachronic

Studies of Languages
,

Pronunciation Lexicon
.

He addressed that
Standardization can help designing better

computational phonology
techniques

and
Computational phonology can help in building

better tools o
r resources that
use the standards

in the conclusion.

He mentioned about speech corpora developed by C
-
DAC Kolkata, C
-
CDAC Noida. He
addressed all

the
requirements specifications used to develop the corpora. He talked about the procedures of recording
and creating databases for corpora. He also talked about the number of speakers,
environment,
speakers age groups,

data coverage
,
parameters to

valida
te the data.


talked about the various requirements occurred in W3C standards like PLS and SSML related to
Indian languages.


Conclusion

1.

Adaption of PLS standard for common
superset
data format for both
TTS and ASR

2.

Need to
create

draft to captu
re phonetic variations among Indian
languages


3.

Need to create documentation contain
ing linguistics requirements and
exceptional examples in Indian languages

4.

Finalization of the metadata and PLS fields used in XML

5.

Need to build a best practices documen
ts which is relevant to Indian
languages



Key issues discussed in Workshop

R
equirements specifications for building both TTS and ASR engines
like:

I.


Recording
-

ITU Standard


16 KHZ PCM

Mono

II.

Coverage of data



-
Desktop environment

-

Noisy environment e.g. Mobile & Speaker in Motion

III.

Number of speakers for TTS and ASR
: 30

IV.

Frequency rate

V.

Environment
: Street, Public Place, Moving Vehicle and
Office/Home

VI.

Metadata: Dublin Core

VII.

Adaption of IPA notations