Study of Nucleus Vowel Duration and

1
Study of Nucleus Vowel Duration and its Role in
Prosody of Bangla
By Rajib Roy, Tulika Basu, Joyanta Basu, Arup
Saha
rajibroy_at_kolkatacdac.in
Centre for Development of Advanced Computing
(C-DAC) Kolkata, India Speech Group
www.kolkatacdac.in
2
Introduction
What is Prosody?
Original speech
Flat synthesized speech

• Prosodic Parameters
• Pitch (F0)
• Duration
• Amplitude
• Pause

The variation of these parameters for a given
dialect depends on sentence, clause, phrase
boundaries, the position of the word, position
and nature of the syllables The study aims to
capture only the details of nucleus vowel
duration from a large corpus of spoken sentences
as we believe that this gives better control for
introduce prosody in synthesized speech.
3
Objective

• The main objective of the study is to find out
  the role of the nucleus vowel duration in the
  prosody of Standard Colloquial Bengali (Bangla).
• The other objective is to study whether nucleus
  vowel duration is related to the phonemic value,
  syllable type and the position of the syllable in
  the word as well as in the clause in Bangla.
• These studies are done in the context of
  introducing naturalness in synthesized speech.

4
Types of syllable

• Syllables are of two types open and closed. If
  the syllable ends in a vowel, it is called an
  open syllable (cv/v) and if it ends in a
  consonant, it is a closed syllable (cvc/vc),
  where v stands for vowels as well as diphthongs
  and c stands for consonants and semivowels.

/ k?k /
/ ?m /
/ o /
/ k? /
Open Syllable (CV)
Closed Syllable (CVC)
5
Nucleus vowel
/ k?k /
Nucleus vowel duration
Nucleus vowel is defined as the steady state of
the vowel along with the two transitions as in
figure. There are seven vowels in Standard
Colloquial Bangla (SCB) /u/, /o/, /?/, /?/, /æ/,
/e/, /i/ (in Bangla )
6
Experimental Data Set

• All together 650 sentences spoken by seven
  informant of both sexes of native SCB are used
  for this study.
• The above data are taken from Bangla Speech
  corpora of C-DAC, Kolkata.
• The data consists of 10185 clauses , 30702 words
  and 58807 syllables .

Meta data of the informants -
7
Experimental Findings and Results
Distribution of Open and Closed Syllables
8
Nucleus vowel duration with respect to syllable
position in a word
9
Nucleus vowel duration for each open syllable
with respect to word position in a clause
10
Nucleus vowel duration for each closed syllable
with respect to word position in a clause
11
Intrinsic Vowel Duration
There exists durational differences for different
vowels due to its articulatory differences
irrespective of its position and context
12
Nucleus vowel duration rules for synthesis
Based on the aforesaid studies a set of rules has
been formed to give durational variation akin to
natural sounding speech to the synthesized output.
Rules

• Take intrinsic vowel duration multiply by 0.99 or
  1.01 if the syllable is closed or open
  respectively.
• Multiply by respective ratio given in Table
  according to the position of the first syllable
  of the word with respect to its position in the
  clause.
• For other syllables multiplication factor is 1.

Ratio of nucleus vowel to intrinsic vowel duration
13
Result of Listening Test

• 15 sentences stimuli were created by replacing
  the nucleus vowel duration of the original
  sentences by the derived duration rule. The
  original and modified sentences are randomly
  mixed up and presented to the subject for
  judgment in 5 value score.
• For this evaluation 5 subjects, 3 male and 2
  female, is selected, their age ranged from 24 to
  50.

14
Result of Listening Test
The total average score for the original
sentences is 3.56 and the modified sentence is
2.94. The average grade difference of less than
one grade is encouraging.
15
Conclusions

• Duration of the nucleus vowel of the 1st syllable
  of consecutive words in the clause decreases. For
  other syllables no such trend is observable.
• Length of the nucleus vowel of the first
  syllable is always longer, irrespective of the
  syllable type.
• It is interesting to note that the occurrence of
  the open syllable is more than two times that of
  the closed in Bangla.
• The average vowel duration of closed syllable is
  greater than that of the open syllable.
• From the study it is observed that the duration
  of high vowels is lesser than the low vowels
  which may be due to some physiological reasons.
• Finally, using the duration rule, TTS output
  becomes-

Flat Synthesized
Duration Modified Synthesized
16
References

• Wen-Hsing Lai Sin-Horng Chen, 2001, A novel
  syllable duration modeling approach for Mandarin
  speech, Proceedings of Acoustics, Speech, and
  Signal Processing, 2001, Volume 1,  pp 93 - 96
  vol.1.
• Uwe D. Reichel, Data-driven Extraction of
  Intonation Contour Classes, pp 240-245, 6th ISCA
  Workshop on Speech Synthesis, Germany, 2007.
• Rao K.Sreenivasa and Yegnanarayana B., Modelling
  Syllable Duration in Indian Languages using
  Neural Networks, pp 313-315, ICASSP, 2004.
• Crystal David, A Dictionary of Linguistics
  Phonetics, Fifth Edition, pp 326, Blackwell
  Publishing, 2003.
• Chatterji Suniti Kumar, The Origin and
  Development of the Bengali Language, pp 402and pp
  279 paragraph 3, 3rd impression, 2002.
• Speech Corpora, CDAC, Kolkata, http//www.kolkatac
  dac.in/html/txttospeeh/corpora/corpora_main/first.
  htm.
• Shyamal Kumar Das Mandal, Datta, Asoke Kumar,
  2007, Epoch synchronous non-overlap-add (ESNOLA)
  method-based concatenative speech synthesis
  system for Bangla, In Proceeding SSW6-2007, pp
  351-355.

17
Thank you