Sat Mar 14 20:08:29 EDT 2009
that recording our own corpus with children would be best as it would
improve recognition efficiency. Also, it was pointed out that we can
choose the corpus that matches a child's vocabulary. As a rough guide,
I have been referring to and extracting corpus from school books (In
India we have the NCERT books available online in PDF format).
> I am interested in applications to foreign language and second
> language learning, including accent reduction.
Our first aim will be to achieve high efficiency in English keeping in
mind accent differences. It's quite feasible to make it work for
this.Then we can move on to other language alphabets but some of them
like Hindi might prove to be more difficult due to extensive use of
diacritics and half-consonants.
When other languages support is added, it becomes a simple matter of
choosing the language and then the child will be given words from that
dictionary and the corresponding model will be used for speech
recognition. This will facilitate foreign language learning. Accent
reduction will automatically happen once the child hears the computer
'espeak' out the words.
More information about the Sugar-devel