17-03-2014, 08:06 PM
Abstract— In this paper we present a rule based concatenative
method for Bangla text to speech conversion. Here our target is
to generate almost natural speech from Bangla PDF text. For
this we have proposed a speech synthesizer, which performs
two major tasks. One is PDF to unicode text conversion and
another is unicode text to speech conversion. We use text
normalization and parsing rules to get syllables which give us
scope to produce natural speech. We have analyzed the
proposed method and an existing method with respect to time
complexity and compared the performance of the methods.
Keywords-PDF; TTS; Unicode.
I. INTRODUCTION
Text-To-Speech (TTS) synthesizer is a computer based
system that should be able to read text from any formatted
documents like PDF, OCR system. Processing of Bangla
language in computers has become more popular and many
innovative applications are being developed by the consistent
and continuous effort of many researchers in Bangladesh [9],
India and in different places in the globe. The TTS for
Bangla uses the widely used Festival TTS engine. Festival is
a concatenative TTS system using diphones [3]. The main
problem in developing applications was the lack of
uniformity. After introduction of unicode in character
encoding the problem was recovered and many applications
like Bangla optical character recognition (OCR for Bangla)
[2], text to speech synthesis, Bangla character coding,
recognition of handwritten in Bangla, etc are being
developed. In this paper, we have proposed a rule based
concatenative synthesis method for generating speech from
Bangla PDF document. We first discuss PDF to unicode text
conversion process and then give an algorithm for the
unicode text to speech conversion. We also discuss the
normalization rules for Bangla pronunciation according to
Bangla academy utcharan ovidhan (pronunciation dictionary)
[10] and parsing rules to generate natural sound. We show
the performance of our unicode text to speech conversion
algorithm with the similar existing method proposed by Firoj
Alam et al [3].
II. RELATED WORK
Several attempts were made in the earlier period, where
different aspects of Bangla Text to Speech system were
covered. In [9] presented some practical applications with
Bangla Text to Speech system using PSOLA (Pitch
Synchronous Overlap Add) technique. It was developed at
France Telecom (CNET) [5]. The Bangla TTS system
proposed by Firoj Alam et al is intelligible, but it has a lack
of naturalness. They did not develop or analyze the system
with document analysis like file format, encoding etc. In [9],
a considerable quantity of work has been done for
developing Bangla TTS. Here diphones are used to build
voice database using ESOLA (Epoch Synchronization and
Overlap Add) technique.
method for Bangla text to speech conversion. Here our target is
to generate almost natural speech from Bangla PDF text. For
this we have proposed a speech synthesizer, which performs
two major tasks. One is PDF to unicode text conversion and
another is unicode text to speech conversion. We use text
normalization and parsing rules to get syllables which give us
scope to produce natural speech. We have analyzed the
proposed method and an existing method with respect to time
complexity and compared the performance of the methods.
Keywords-PDF; TTS; Unicode.
I. INTRODUCTION
Text-To-Speech (TTS) synthesizer is a computer based
system that should be able to read text from any formatted
documents like PDF, OCR system. Processing of Bangla
language in computers has become more popular and many
innovative applications are being developed by the consistent
and continuous effort of many researchers in Bangladesh [9],
India and in different places in the globe. The TTS for
Bangla uses the widely used Festival TTS engine. Festival is
a concatenative TTS system using diphones [3]. The main
problem in developing applications was the lack of
uniformity. After introduction of unicode in character
encoding the problem was recovered and many applications
like Bangla optical character recognition (OCR for Bangla)
[2], text to speech synthesis, Bangla character coding,
recognition of handwritten in Bangla, etc are being
developed. In this paper, we have proposed a rule based
concatenative synthesis method for generating speech from
Bangla PDF document. We first discuss PDF to unicode text
conversion process and then give an algorithm for the
unicode text to speech conversion. We also discuss the
normalization rules for Bangla pronunciation according to
Bangla academy utcharan ovidhan (pronunciation dictionary)
[10] and parsing rules to generate natural sound. We show
the performance of our unicode text to speech conversion
algorithm with the similar existing method proposed by Firoj
Alam et al [3].
II. RELATED WORK
Several attempts were made in the earlier period, where
different aspects of Bangla Text to Speech system were
covered. In [9] presented some practical applications with
Bangla Text to Speech system using PSOLA (Pitch
Synchronous Overlap Add) technique. It was developed at
France Telecom (CNET) [5]. The Bangla TTS system
proposed by Firoj Alam et al is intelligible, but it has a lack
of naturalness. They did not develop or analyze the system
with document analysis like file format, encoding etc. In [9],
a considerable quantity of work has been done for
developing Bangla TTS. Here diphones are used to build
voice database using ESOLA (Epoch Synchronization and
Overlap Add) technique.