word boundary- hypothesisation in hindi speech - Speech and ...

purpose of the symbol-to-text conversion system is to make the symbol string readable,
by providing the missing word bou,ndaries. The text can be further corrected, if
necessary, using the higher level knowledge sniirces such as syntax and semantics to
make it more meaningful.
In summary, the following are the advantages of the word boundary
hypothesisation:
1. The complexity of lexical matching involved in large vocabulary speech recognition
can be significantly reduced.
2. Unknown words can be handled.
3. If most of the word boundaries can be hypothesised, a useful speech-to-text
conversion system can be developed, with only a speech signal-to-symbol converter and
a word boundary hypothesiser.
It is interesting to note that a meaningful text with word boundaries can be read
easily, even with some errors in characters and in word boundaries (see Fig.l.1 for
illustration). Thus word boundary hypothesisation plays a crucial role in producing a
readable output from a speech-to-text conversion system. But continuous speech does
not contain any direct clues, such as pauses, to word boundaries. However, it is
interesting to note that there are several language features which can be exploited for
hypothesising word boundaries. Since the original input is speech signal, one can also
exploit speech related clues for word boundary hypothesisation.
The objective of this thesis is to establish the significance of word boundary
hypothesisation in speech recognition and to demonstrate that language and speech
related clues do exist, which can be effectively used to hypothesise word boundaries. It
is interesting to note that even a partial success in word boundary hypothesisation using
these clues would generate a text which is significantly better than a text without word
boundaries, from a readability point of view. Moreover, such a text with a few word