Professional Documents
Culture Documents
Marathi Speech Synthesis A Review
Marathi Speech Synthesis A Review
Volume: 3 Issue: 6
ISSN: 2321-8169
3708 - 3711
_______________________________________________________________________________________________
Kavita Waghmare1
Abstract This paper seeks to reveal the various aspects of Marathi Speech synthesis. This paper has reviewed research development in the
International languages as well as Indian languages and then centering on the development in Marathi languages with regard to other Indian
languages. It is anticipated that this work will serve to explore more in Marathi language.
Keywordsspeech synthesis; festival; articulatory synthesis; formant synthesis;concatenative synthesis, Hidden Markov Model.
__________________________________________________*****_________________________________________________
I.
INTRODUCTION
_______________________________________________________________________________________
ISSN: 2321-8169
3708 - 3711
_______________________________________________________________________________________________
This method applies the parameters such as fundamental
frequency, voicing and noise levels over a point of time to
make a waveform of artificial speech.
C. Hidden Markov Model
The Hidden Markov Model (HMMs) is widely-used statistical
models to characterize the sequence of speech spectra and have
successfully been applied to speech recognition and speech
synthesis systems. This system simultaneously models,
spectrum, excitation, and continuance of speech using contentdependent HMMs and generates speech waveforms. HMM
creates stochastic models from known utterances and compares
the probability that the unknown utterance can be engendered
by each model. It also includes various methods for text to
voice communication such as Dynamic Features (Delta and
Delta-Delta parameters of Speech). HTK (Hidden Markov
Model Toolkit) is a software toolkit for handling HMMs
applications that employ HMMs for speech synthesis. This
technique uses very less memory, but consumes large CPU
resources. This approach gives good prosody features with
natural sounding language.
D. Concatenative Synthesis
Concatenative synthesis is the most uncomplicated method to
synthesize the speech which is got by concatenating the
different sentences, words, syllables, phones, iPhones, and
triphone. These are already stored to get the desired output
language. It requires large databases, sometimes it is quite
impossible to store. More natural sounding speech is produced
by this technique. Sometimes, the difference between natural
variations and the nature of the automated techniques for
segmenting the waveforms may result in audible glitches in the
production. There are three sub-types of Concatenative
synthesis:
1) Unit selection Synthesis or Corpus based synthesis: Unit
selection synthesis is also referred as corpus based synthesis. It
uses large database. During database creation, each recorded
utterance is segmented into some individual phones, syllables,
morphemes, words, phrases, and sentences. An index of the
units in the speech database is then made based on the
segmentation and acoustic parameters such as fundamental
frequency, pitch, duration, the status of the syllable and
previous and next phones. This method provides naturalness in
output speech as compared to other techniques.
2) Diphone Synthesis: Diphone synthesis techniques require
less database as compared to the unit selection synthesis. It uses
two adjacent phones to make the speech waveform. But this
techniques suffers through the problem of coarticulation.
3) Domain specific synthesis: Domain specific synthesis is
associated to the particular field. In this database consists of
_______________________________________________________________________________________
ISSN: 2321-8169
3708 - 3711
_______________________________________________________________________________________________
[6] [7] [8]. C-DAC is the institution which has been working in
the area of speech synthesis since 25 years. This institute has
developed text-to-voice communication system in Hindi,
Malayalam, Bangla, Mizo and Nepali. They have developed
ESNOLA based Bangla synthesis techniques. It is called as
BANGLA VAANI. Other systems are Mizo Text reading
system, Mozhy TTS. [8] [9] [10] [11] [12][13][14].
A. Applications available in International languages
Classic text to speech engine: It is a Text To Speech Engine
from SVOX, in combination with 40+ male/female voices in
more than 25 languages that allows to read aloud texts from ebook, navigation, translation and other apps. The languages in
which it is available are Arabic (male), Australian English
(female), Brazilian Portuguese (female), Canadian French
(male/female), Cantonese (female), Czech (female), Danish
(female), Dutch (male/female), Finnish (female), French
(male/female), German (male/female), Greek (female),
Hungarian (female), Italian (male/female), Japanese (female),
Korean (female), Mandarin (female), Mexican Spanish
(male/female), Norwegian (female), Polish (female),
Portuguese (male/female), Russian (male/female), Slovak
(female), Spanish (male/female), Swedish (female), Thai
(female), Turkish (male/female), UK English (male/female),
US English (male/female).
a) IVONA: It is a TTS system available in 17 languages. It
gives natural sounding and more accurate voices. It is
compatible with Windows, Unix, Android, Tizen, iOS based
systems. The compatible languages are American, Australian,
British,
Welsh,
German,French,Castilian,Icelandic,Italian,Canadia,Dutch,Euro
pean,Brazilian,Polish,Romanian,Russian,Danish.
b) CereVoice Engine: It gives support to 9 languages. It can
be easily deployed with any variety of English voices. It is
available in English, French, Spanish, Italian, German,
Portuguese, Japanese, Dutch, and Catalan.
c) eSpeak: It supports 51 languages over the world. It lets in
different voices, whose features can be changed. It partially
supports SSML (Speech Synthesis Markup Language).The
languages supported by eSpeak are Afrikaans, Albanian,
Aragonese, Bulgarian, Danish, Dutch, Cantonese, Catalan,
Croatian, English, Esperanto, Estonain, Farsi, Finnish, French,
Georgian, German, Greek, Hindi, Hungarian, Icelandic,
Indonesian, Irish, Italian, Kannada, Kurdish, Latvian, Lojban,
Malayalam, Malaysian, Nepalese, Norwegian, Polish,
Portuguese,Punjabi,Romanian,Russian,Serbain,Slovak,Spanish,
Swahili,Swedish,Tamil,Turkish,Vietnamese.
d) Google Text-to-Speech: This TTS system supports 15
languages. It is a low level speech synthesis. The languages
_______________________________________________________________________________________
ISSN: 2321-8169
3708 - 3711
_______________________________________________________________________________________________
component of a larger speech application. It can be used to
simply synthesize text file.
VI. CONCLUSION
Speech synthesis is a continuously developing technology
since many years. This paper has presented various aspects of
speech synthesis such as methods, tools, techniques,
applications currently available in various languages. It is seen
that the work in the area of Marathi speech synthesis is very
less as compared to other Indian languages and it can be
explored more.
[4]
[5]
[6]
[7]
[8]
[9]
Mr.S.D.Shirbahadurkar,
Dr.D.S.Bormane,Speech
synthesizer Using Concatenative Synthesis Strategy for
Marathi Language,International Journal of Recent Trends
in Engineering,Vol 2,No.4,November 2009.
[10]
https://www.ivona.com/
[11]
https://www.cereproc.com/en/products/sdk
[12]
http://espeak.sourceforge.net/
[13]
https://apps.mgov.gov.in/descp.do?appid=527
[14]
http://sourceforge.net/projects/as-peak/
REFERENCES
[1]
[2]
[3]
3711
IJRITCC | June 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________