Just to say that I have been doing some work on Fx and MBROLA.
1. I have fixed some bugs in uttalign
2. I have written a program 'uttfx' that fits lines to the fundamental
frequency of an utterance and transfers the FX midpoint (FXMID) and
Fx gradient (FXGRD) onto the prosodic structure.
3. I have written a program 'uttmbrol' which outputs MBROLA compatible
.pho files from our XML annotations. It uses a cubic curve to link
the FXMID values of adjacent segments, respecting the FXGRD values
at start and end. It seems to produce acceptable MBROLA output, but
I don't know yet whether the Fx modelling is accurate enough.
I am going to look at prosody manipulation next. The message below
from Gerard Bailly is interesting as there will be an evaluation of
various codecs within COST 258 and I would like to take part.
It would be useful for us to think about some standard phrases for
testing prosody manipulation: pairs of contrasting sentences and
a neutral version perhaps.
Regards
Mark
--------------------------------------------------------------
>Sender: bailly@icp.inpg.fr
>Date: Tue, 28 Jul 1998 18:32:31 +0200
>From: Gerard Bailly <bailly@icp.inpg.fr>
>Organization: ICP
>X-Mailer: Mozilla 3.04Gold (X11; I; Linux 2.0.33 i686)
>To: Ove Andersen <oa@cpk.auc.dk>, Gerard Bailly <bailly@icp.grenet.fr>,
> ER Banga <erbanga@tsc.uvigo.es>, Andy Breen <apb@dwarf.bt.co.uk>,
> Mike Eddington <mde@dwarf.bt.co.uk>,
> Gudrun Flach <flach@eaksw2.et.tu-dresden.de>,
> "Mr. Horak" <horak@ure.cas.cz>,
> Mark Huckvale <mark@phonetics.ucl.ac.uk>,
> Steve Isard <stepheni@cstr.ed.ac.uk>, Gernot Kubin
<g.kubin@ieee.org>,
> "Luis.C. Oliveira" <Luis.Oliveira@inesc.pt>,
> Georg Ottesen <Georg.E.Ottesen@informatics.sintef.no>,
> Thomas Portele <tpo@ikp.uni-bonn.de>,
> Beat Pfister <pfister@tik.ee.ethz.ch>,
> "Mr. Pribil" <pribil@ure.cas.cz>, Jacques Terken <terken@ipo.tue.nl>,
> Robert Vich <vich@ure.cas.cz>,
> Brigitte Zellner <brigitte.zellner@imm.unil.ch>,
> Eric Keller <eric.keller@imm.unil.ch>
>Subject: Evaluation of coders
>
>Dear evaluators of coders,
>
> In my last mail, I proposed to evaluate coders by
>using prosodic transplantation. I have got
>nice and encouraging personal reactions. Thank to
>all of you. In order to illustrate my proposal,
>I have added on the cost258 web server:
>http://www.icp.inpg.fr/cost258/signal_processing/ressources.html
>some information and samples:
>- the source 6_ym_NT_115.wav is at 16kHz
>- the target 6_ym_DI_115.wav is at 10kHz
> I give here the audio file just for reference. The coder only uses
> the prosodic characteristics of it:
> melody (specified by 6_ym_DI_115.pca)
> segment durations and energies (specified by 6_ym_DI_115.seg)
>- the file 6_ym_DI_115_6_ym_NT_115.wav was obtained by transplanting
> the source according to prosodic characteristics of the target
> using TDPSOLA.
>
>... TDPSOLA is impressive.. isn't it?
>
>I will add very quickly samples from other ICP coders under development.
>
>I do have more (source,target) couples .. more difficult ones!!! And I
>am
>ready to put more of them on the web. I thus suggest that people
>attending Vigo come with transplanted sounds with their own coders.
>I will be happy to add them also in the server.
>
>I need reaction on this proposal. Could we (you) make it til Vigo?
>
>You work too hard.
>
>Friendly
>Gerard
>____________________________________________________________
>NOTE THE NEW EMAIL ADDRESS
>____________________________________________________________
>Gerard Bailly (bailly@icp.inpg.fr)
>Institut de la Communication Parlee, URA CNRS 368
>INPG/Universite Stendhal
>46, av. Felix Viallet. 38031 GRENOBLE CEDEX. FRANCE
>Tel: (33) 04 76 57 47 11 Fax: (33) 04 76 57 47 10
>____________________________________________________________
>
>