Re: [orca-list] New eSpeak version soon, please send corrections

[Jonathan Duddington <jonsd jsd clara co uk>]

On 18 Nov, Krishnakant <hackingkk gmail com> wrote:

> > I think eSpeak makes many mistakes about:
> > 1.  Which syllable of a word should have the main stress.
> > 2.  When the "inherent vowel" "a" should be silent, and when it
> > should be spoken.
> > 3.  Nasal vowels.

> point number 2 is valid but espeak does pritty well with the rest.

English and other languages have similar problems.  To fix this, we
will need to add many more rules and exceptions to the hi_rules and
hi_list files.

Compare the size of the English en_rules and en_list files, although
I'm sure English spelling-to-sound translation is more difficult than
Hindi.

> I think the problem with espeak is that it talks hindi in a british
> axent which is obvious and perhaps unchangable.

No, it can be changed.  But you need to listen carefully and identify
exactly what sounds wrong.

Perhaps some of the vowels or consonants have an "English" sound?

Another obvious difference is the intonation.  This is the way in which
the pitch of the voice changes during a sentence (like a "song"). 
Indian languages have a different intonation from English.

It would be an interesting project for someone to write an intonation
algorithm which is suitable for Hindi.

The problem is this:
You have a sequence of syllables which make up a sentence.  The
stressed syllable of each word is marked.  You need to give each
syllable (vowel) a starting pitch and an end pitch.

The intonation may be different for different types of sentence/clause,
depending whether it ends in full-stop, question mark, exclamation, or
comma (or their equivalent in Devanagari script).