Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Automatic Glottal Closure Instant Identification.(GCI)

1 view
Skip to first unread message

rajesh

unread,
Apr 28, 2008, 2:22:57 AM4/28/08
to
Hi,
I have developed an algorithm for automatic glottal closure
instant identifiaction from the speech signal. It is based on a
combination of wavelet based matched filtering,some kind of dynamic
programming and epoch filtering. It can identify the GCI's very
accuraltely.

Please suggest me any database (with laryngograph results)(freely
available) for testing the same.

GCI identification has applications in speaker verification. Anyone
interested in collaboration
please contact me.


Rajesh.D

GB

unread,
Apr 28, 2008, 10:40:21 AM4/28/08
to

"rajesh" <getra...@gmail.com> wrote in message
news:1e9dd33e-18f8-4b14...@f24g2000prh.googlegroups.com...

I'm just new to all this. I (apparently wrongly) assumed that this would be
a routine problem that was solved many years ago. It is pretty obvious from
looking at a wide-band spectrograph when the glottal closures occur, and I
assumed it was equally easy to come up with a suitable algorithm.

A quick google search comes up with the following reference:
http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/10376/4032760/04032783.pdf?isnumber=4032760&prod=JNL&arnumber=4032783&arSt=34&ared=43&arAuthor=Patrick+A.+Naylor%3B+Anastasis+Kounoudes%3B+Jon+Gudnason%3B+Mike+Brookes

This mentions the following databases (in answer to your question):
'Results are presented for the APLAWD and SAM databases for which 95.7% and
93.1% of GCIs are correctly identified'


David Gelbart

unread,
Apr 28, 2008, 1:44:25 PM4/28/08
to
"Please suggest me any database (with laryngograph results)(freely
available) for testing the same."

MOCHA-TIMIT?

http://www.cstr.ed.ac.uk/research/projects/artic/mocha.html

David Gelbart

unread,
Apr 28, 2008, 2:48:21 PM4/28/08
to
"Please suggest me any database (with laryngograph results)(freely
available) for testing the same."

MOCHA-TIMIT?

http://www.cstr.ed.ac.uk/research/projects/artic/mocha.html

rajesh

unread,
Apr 29, 2008, 2:52:30 AM4/29/08
to
On Apr 28, 7:40 pm, "GB" <NOTsome...@microsoft.com> wrote:
> "rajesh" <getrajes...@gmail.com> wrote in message

>
> news:1e9dd33e-18f8-4b14...@f24g2000prh.googlegroups.com...
>
>
>
> > Hi,
> > I have developed an algorithm for automatic glottal closure
> > instant identifiaction from the speech signal. It is based on a
> > combination of wavelet based matched filtering,some kind of dynamic
> > programming and epoch filtering. It can identify the GCI's very
> > accuraltely.
>
> > Please suggest me any database (with laryngograph results)(freely
> > available) for testing the same.
>
> > GCI identification has applications in speaker verification. Anyone
> > interested in collaboration
> > please contact me.
>
> > Rajesh.D
>
> I'm just new to all this. I (apparently wrongly) assumed that this would be
> a routine problem that was solved many years ago. It is pretty obvious from
> looking at a wide-band spectrograph when the glottal closures occur, and I
> assumed it was equally easy to come up with a suitable algorithm.
>
> A quick google search comes up with the following reference:http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/10376/4032760/0...

>
> This mentions the following databases (in answer to your question):
> 'Results are presented for the APLAWD and SAM databases for which 95.7% and
> 93.1% of GCIs are correctly identified'

Hi GB .,
thank you.

here is a generalization of the problem. please let me know if i
have made any mistakes.


Consider f(t) and its kth derivative fk(t). fk(t) is continous in
(a,b) for k = 1,2,3...n-1.
fn(t) is also continous in (a,b) except at t=t0.

find g(t) such that g(t) has a local maxima in (a,b) at t=t0.

In the paper you have mentioned, http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/10376/4032760/0...
they are not addressing the above problem.

Please give me your comments on this.

My algorithm operates on narrowband speech(8Khz) and can detect GCI's
upto a 0.5 sample accuracy which ofcourse has to be tested on a
database.

regards
Rajesh.D

rajesh

unread,
Apr 29, 2008, 2:54:11 AM4/29/08
to

Hi David,

thanks for your suggestion, i am hopeful that it is
useful for me.


regards
Rajesh.D

rajesh

unread,
May 6, 2008, 5:21:53 AM5/6/08
to

Thanks for your suggestion. I have used the database successfully.
I am getting a result of identification rate close to 90%.

rajesh

unread,
May 6, 2008, 5:53:18 AM5/6/08
to

> This mentions the following databases (in answer to your question):
> 'Results are presented for the APLAWD and SAM databases for which 95.7% and
> 93.1% of GCIs are correctly identified'

I have personally written to the authors of the above mentioned
papers and requested for
the database. Unfortunately the APLAWD database copyright owners have
stopped trading.

Mark Huckvale

unread,
May 9, 2008, 5:31:18 PM5/9/08
to
Rajesh

Paul Bagshaw's database:

http://www.cstr.ed.ac.uk/research/projects/fda/

has Laryngograph waveform files that you could use to generate reference
pitch marks, as does the Keele Pitch database:

http://www.liv.ac.uk/Psychology/hmp/projects/pitch/speech/keele_pitch_database.html

If you try out a new algorithm, be sure to compare it against an
existing standard program. Recently I have been using Praat for
generating pitch marks for some pitch synchronous analysis and it seems
as good as any.

Regards

Mark Huckvale


--
Mark Huckvale, Director MSc Speech and Hearing Science
Phonetics and Linguistics, University College London
www.phon.ucl.ac.uk

rajesh

unread,
May 19, 2008, 3:44:57 AM5/19/08
to

Thank you Mark.

here are some more databases with EGG recordings that i have used.

http://www.festvox.org/cmu_arctic/

http://www.festvox.org/dbs/dbs_kdt.html

Rajesh.D

0 new messages