Study on phonetic context of Malay syllables towards the development of Malay speech synthesizer
Loading...
Date
2007
Authors
Samsudin, Nur Hana
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Speech synthesizer has evolved from parametric speech synthesizer (articulatory and
formant synthesizer) to non-parametric synthesizer (concatenative synthesizer).
Recently, the concatenative speech synthesizer approach is moving towards corpusbased
or unit selection technique. In this approach, the pre-recorded speech segments
which are to be used in the synthesizer are stored exactly as how it is recorded.
Additional information of the speech waveform is attached to the sound to provide proper
annotation of the speech waveform.
However, annotations of the speech waveform remain as a loose standard. What should
be annotated and how a unit selection process is carried out rely heavily on the
developer, and target language, as well as the target domain of the synthesizer usage.
Features used to represent pre-recorded speech are varied and treated as language
dependent.
Until this thesis is written, we are still unaware of any study related to what speech
features should be made available in a Malay speech corpus for a Malay speech
synthesizer.
This thesis addresses the issues of producing a more natural sounding speech
synthesizer for Malay. We focus on high level representation of speech features which is
the phonetic context of the speech to utter. We conducted an observation on the effect
of phonetic context to the quality of concatenated speech. Our hypothesis is to show
that, to get the best concatenative speech result, we have to find similar or closest match
of phonetic context, between the recorded utterance and target utterance. Hypothetically
also, the output quality of this selective method will be better than when we select a
segment in random.
We also proposed a template which will guide the system to select the best candidate of
the speech segment to produce a better synthesized speech for longer utterance. This is
because, our study covers only the phonetic aspect of the speech and we did not
discussed on other aspects of speech in detail. The template will detail out additional
criterion which need to be followed during selection of unit to be concatenated.
At the end of the research, we are able to give out the performance and preference
rating of the concatenated speech which is based on phonetic context. Finally, we
presented the future work to further improve Malay speech synthesizer.
Description
Master
Keywords
Science computer , phonetic context , Malay syllables , Malay speech synthesizer