hi
i am doing a project on speech recognition using
nueral networks.
front end extraction is done by mfcc.
how could i use praat to extract mel frequency
cepstral coefficients(mfcc) ??

__________________________________________________
Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
http://mailplus.yahoo.com

Hello everybody,

I need to use the source-filter synthesis of Praat, to modify the formants of
speech (word and phrases) and to do resynthesis.
I have some questions:

1) After I have obtained the formants from a Sound object (I use "To formant
(burg)... 0.005 5 0.025 50"), I "Write to text file..." these values and I find
that for some frames there aren't all five formants (...and I need these values,
that are associated to resonancies of vocal tract).
If I use "To formant (keep all)..., for some frames I find 6 formants...and it
is difficult to modifies always the same formant (i.e. F2).
If I use "To formant (sl)... 0.005 5 0.025 50" (i.e.Split Levinson's algorithm)
I always find the requested number of formants in every frame, but in Praat
manual is written that Burg's algorithm "yelds much more reliable formant
values..."
How can I do? If I set correctly the maximum formant value for my male speaker,
can I use Burg's algorithm to obtain always the five values of formants?

2)When I use "To formant (burg)...", can I use a Sound object with any sampling
frequency (because it is resampled from algorithm implemented  in Praat,
according to maximum formant value) or it is necessary that I do first my
"manual" resampling with Praat?

3) Is it possible to assign a different color to each formant in the "Edit
window" of a generic Sound object? I think this is necessary in order to
distinguish better F1, F2, F3, F4 and F5.

Thank you very much for an answer.

Best Regards.

**************************************************************
Ing. Francesco Marzulli
c/o Politecnico of Bari
       Department of Electrical & Electronic Engineering
       Via Orabona, 4 - 70125 Bari, Italy
     E-Mail : marzulli@...
**************************************************************





[Non-text portions of this message have been removed]

I have been a CSL user and have recently been directed toward PRAAT
for a dissertation project. CSL makes alot of fuss over their
hardware being far superior to standard PC hardware when converting a
signal. Is there a qualitative difference in the ultimate signal?
Also, it seems that this would differ as a function of specific
hardware. I am ordering a new computer that can be spec'd with
features I need. Is there a particular set up that is desireable for
using PRAAT? What would you suggest?

hi,

I'm having some trouble using the Get minimum and maximum pitch
commands. The problem is that I get different results depending on
whether I (a) "edit" a Sound+TextGrid containing several intervals,
select the interval I'm interested in and query it, or (b) "extract"
the intervals, then "edit" the interval, select it in its entirety
and query it. As far as I can tell, the results should be identical,
but the values for min and max pitch (and mean, for that matter)
differ, depending on which I do.
Can anyone help?  - Many thanks in advance!

Susanne

You could get a digital sound card or an external audio capture box.
Edirol (www.edirol.com) is a good place to look for the latter.  They
have in mind music applications, so I think even their portable models
would be adequate for linguistic phonetics research, as well as affordable.

-K.

praat-users@... wrote:

> To Post a message, send it to:   praat-users@eGroups.com
> To Unsubscribe, send a blank message to: praat-users-unsubscribe@eGroups.com
> To consult archives :          http://egroups.com/list/praat-users/
>
> ------------------------------------------------------------------------
>
> There is 1 message in this issue.
>
> Topics in this digest:
>
>       1. Hardware advice
>            From: "whi_me2002" <irwinwh@...>
>
>
> ________________________________________________________________________
> ________________________________________________________________________
>
> Message: 1
>    Date: Tue, 03 Dec 2002 21:21:50 -0000
>    From: "whi_me2002" <irwinwh@...>
> Subject: Hardware advice
>
> I have been a CSL user and have recently been directed toward PRAAT
> for a dissertation project. CSL makes alot of fuss over their
> hardware being far superior to standard PC hardware when converting a
> signal. Is there a qualitative difference in the ultimate signal?
> Also, it seems that this would differ as a function of specific
> hardware. I am ordering a new computer that can be spec'd with
> features I need. Is there a particular set up that is desireable for
> using PRAAT? What would you suggest?
>
>
>
> ________________________________________________________________________
> ________________________________________________________________________
>
>
>
> Your use of Yahoo! Groups is subject to
http://uk.docs.yahoo.com/info/terms.html
>
>
>
>

Does anyone know what the difference between the 6 jitter
measurements is?  I have the same question for the 3 varying shimmer
measurements.

Finally, what exactly does mean absolute slope calculate?  My hope
is that it is a measure of Fo velocity or how rapidly Fo changes.
Does anyone know if this is the case?

Thanks for any insight you might have.

Cheers,
Michael

At 17:09 +0000 6-12-2002, leshsakte wrote:
>I'm having some trouble using the Get minimum and maximum pitch
>commands. The problem is that I get different results depending on
>whether I (a) "edit" a Sound+TextGrid containing several intervals,
>select the interval I'm interested in and query it,

this is usually the correct way. In this case, Praat can reliably measure pitch
values throughout the selected interval, even at the edges, because some of the
waveform outside the interval is known as well.

When Praat computes a pitch value at a given timepoint, it takes into account 20
milliseconds of the waveform to the left of this point and 20 milliseconds of
the waveform to the right of this point (the number of milliseconds depends on
the "Minimum pitch" setting). The pitch value at the left edge (beginning) of an
interval is therefore computed on the basis of a part of the waveform that is
outside the interval. The same goes for the right edge.

> or (b) "extract"
>the intervals, then "edit" the interval, select it in its entirety
>and query it.

In this case, Praat can take into account only the waveform inside the interval.
For Praat, this means that it can compute no pitch values near the edges. If you
"edit" a short interval, you see that the pitch curve does not extend all the
way to the edges: this is not because Praat thinks that the sound is voiceless
in these areas, but because Praat does not have enough information to compute a
reliable pitch value there.

>As far as I can tell, the results should be identical,
>but the values for min and max pitch (and mean, for that matter)
>differ, depending on which I do.

So now you understand why the results are different. If you are interested in
local properties of the global pitch contour, you may use the first method. If
you want to exclude information about the waveform outside the interval, you may
use the second method, thereby taking the minimum and maximum of the pitch
values near the centre of the interval only, but if the interval is very short
(e.g. less than 40 milliseconds), Praat will compute no pitch value at all
anywhere inside the interval.
--

Paul Boersma
Institute of Phonetic Sciences, University of Amsterdam
Herengracht 338, 1016CG Amsterdam, The Netherlands
http://www.fon.hum.uva.nl/paul/
phone +31-20-5252385

At 12:00 +0100 3-12-2002, F.Marzulli wrote:
>How can I do? If I set correctly the maximum formant value for my male speaker,
can I use Burg's algorithm to obtain always the five values of formants?

You cannot. Five formants means "ten poles", and sometimes those poles will
suggest formants around 0 Hz, which nobody would like to consider to be an F1.
That's why you get fewer than five formants in some frames. Also, sine waves
simply have a single formant, so a really good formant analysis algorithm would
have to find only one formant there. In silent signals, Praat will find no
formants at all.

>2)When I use "To formant (burg)...", can I use a Sound object with any sampling
frequency (because it is resampled from algorithm implemented  in Praat,
according to maximum formant value) or it is necessary that I do first my
"manual" resampling with Praat?

no manual resampling is necessary.

>3) Is it possible to assign a different color to each formant in the "Edit
window" of a generic Sound object? I think this is necessary in order to
distinguish better F1, F2, F3, F4 and F5.

You can probably judge better than Praat whether a red dot at 1000 Hz should be
considered an F1 or an F2... Well, I can imagine that having five different
colours is a nice option. Once Praat can understand five perceptually
contrasting variants of red, I will program this.
--

Paul Boersma
Institute of Phonetic Sciences, University of Amsterdam
Herengracht 338, 1016CG Amsterdam, The Netherlands
http://www.fon.hum.uva.nl/paul/
phone +31-20-5252385

Hello,

In case anyone else has this or a similar problem in
the future:

I figured out the problem with Praat reading [Olympus
W-10 Voice Recorder]-created WAV files.  These files
are compressed, and it was a matter of finding
software to decompress them so Praat can read them.  I
recommend a simple freeware called dBpowerAMP Music
Converter, where you just click a couple of times and
it's converted.

Happy Holidays,
Lauren Hall-Lew

__________________________________________________
Do you Yahoo!?
New DSL Internet Access from SBC & Yahoo!
http://sbc.yahoo.com

Hi all.

The path problem has been solved. I discovered the environment$ hash.
Now for a different problem:
Is it possible to wait for something before continuing the execution of
the script.

My example is that I would like to wait for an editor to finish before
saving the resulting TextGrid. (Don't want to save the Grid before I've
had time to put something in it..)

pause is my best suggestion.
Improvements anyone?

/Fredrik

--
-----------------------------------------------------------------
Fredrik Karlsson, Research Student

Department of Philosophy and Linguistics
Umeå University
S-901 87 UMEÅ
SWEDEN
Tel: +46 90 786 56 84   Web: http://www.ling.umu.se/~fredrik/
Fax: +46 90 786 63 77   Email: fredrik@...
PGP-fingerprint:2DD5 54A1 4555 6D00 F386  C269 C430 0C0F 1A53 49B9
-----------------------------------------------------------------

Hello,
Many times I need to load a lot of little audio files from the disk
(p.e. to convert from a type to another one). In this case, I cannnot
get several files simultaneously and I must to read them one to one.

Is it not possible to read (Read/Read from file..) and get several
files at the same time? (without using script language).

On the other hand, in script language I not find a general order to
read for example all the files of a directory given.

Anyone can help me? (I would save a lot of lost time)

Julio Gonzalez PhD
University Jaume I. Castellon (Spain)

Here is a script I use.  I plant it in the Read menu.

form Multiple File Loader
    text Path_To_Files C:\
endform
directory$ = "'Path_To_Files'"
Create Strings as file list... list 'directory$'\*.wav
numberOfFiles = Get number of strings
for ifile to numberOfFiles
     select Strings list
     filename$ = Get string... ifile
     Read from file... 'directory$'\'filename$'
endfor
select Strings list
Remove



>>> "jgonza27es" <gonzalez@...> 01/23/02 05:48AM >>>
Hello,
Many times I need to load a lot of little audio files from the disk
(p.e. to convert from a type to another one). In this case, I cannnot
get several files simultanesly and I must to read them one to one.

Is it not possible to read (Read/Read from file..) and get several
files at the same time? (without using script language).

On the other hand, in script language I not find a general order to
read for example all the files of a directory given.

Anyone can help me? (I would save a lot of lost time)



To Post a message, send it to:   praat-users@eGroups.com
To Unsubscribe, send a blank message to: praat-users-unsubscribe@eGroups.com
To consult archives :          http://egroups.com/list/praat-users/


Your use of Yahoo! Groups is subject to http://uk.docs.yahoo.com/info/terms.html

Hi!
     I need to integrate some scripts thata I´ve made with Praat into
a bigger C programm. Is it possible?

     Thank you,
     Rocio.

Dear Praat users,

I'm working with vowel recognition, but I'm not familiarized with "praat".
I would like to know a easy way to:
- Get the average F1 and F2 from a selected vowel.
- Change F1 or F2 values in a determined vowel.
- Make synthetics vowels only with F1 and F2 given values.

Sorry for the bad English.

Thanks in advance for any suggestion.

Ricardo A. H. Bion,

Florianópolis-Santa Catarina-Brazil.
------------------------------------


_________________________________________________________________
Chegou o novo MSN Explorer. Instale já. É gratuito!
http://explorer.msn.com.br

Hi,

     Is it possible to merge tiers at a specific time?  I have two
text grids and I need to combine them.  I figured out how to combine
the waves (making a sound chain).  But the tiers are a different
matter.  When I merge the tiers together they sit one on top of the
other, and I need them side by side.

Can anyone help?

Jennifer Lukas
Graduate Student
York University
Toronto, Canada

Dear Paul and others!

Thanks for the information, and for the suggestions.  Unfortunately, I am
running Praat on a machine that does not have a lot of
memory, and I have a large number of sound files to go through, so I can't keep
a lot of files open at the same time.  What I've come up
with is this:  I added two buttons to the dynamic menu, and I added a pause to
the form with a loop so that it will pause after drawing
the Formant object.  While it's paused, you can click on the "Formant OK" button
to save the selected object  or the "Bad Formant"
button to add the name of the selected object to a text file.  Then, when you
click on "continue", all the objects are removed except
the Strings list (to free up memory), and we go on to the next sound.  I am
using selected$ ("Formant") to get the name, so it all
works out OK as long as you don't click on some other object in the list while
the pause is in effect.  Seems to work OK.  :-)

Now -- here's my attempt at an argument for why making Praat scripts have
wizard-like abilities would be useful.  Suppose for some
reason that you wanted to measure formants by hand.  Maybe this is because you
are teaching a course in phonetics and you want the
students to learn to read a spectrogram; maybe it is because you are in one of
those situations where you can see a formant but can't
get it to show up in the formant object.  (You may, perhaps, have a handy list
of sound files where precisely this problem is
ocurring...)  In this situation, it would be nice to be able to cycle through
the files to be measured, and for each one have a handy form
pop up asking you to enter your hand measurements.  This would be handy because,
as I understand it, there is no existing way to log
such a measurement.  And even if I am overlooking an easy way to do this without
forms, I think that there would still be some neat
things you could do with this capability.  For example, you could have a lab
assignment where students come in and run the
appropriate script.  The script loads a sound object, and opens it in the sound
editor, making sure that "view formants" is turned off,
and that the cursor is placed at an appropriate place (say the middle of some
vowel).  Then a form pops up, asking the student to enter
F1, F2, and F3 at the cursor location as measured by hand.  Then, you could just
save the measurements to some file with the student's
name (a nifty type of quiz), or you could have the script display the "correct"
values alongside the student's measurements for instant
feedback.  In any case, you would then move on to the next sound file, or the
next vowel in the same sound file.  Well, that's the idea I
came up with.  What do you think?

Thanks again,

Katherine.

I am doing a pitch analysis and am editing some of the
pulses ("pitch markers") in the manipulation window.
I then want to generate new pitch points based on
these new pulses.

When I use the command <pitch>, <add pitch point at
time slice> I get one pitch point that represents the
average pitch for that time period, while the original
analysis generated 15 pitch points for the same time
period.

Is there any way to have Praat recalculate the pitch
points for a time period like it does in the original analysis??

__________________________________________________
Do You Yahoo!?
Yahoo! Games - play chess, backgammon, pool and more
http://games.yahoo.com/

I have a question to do: I want to change the pitch of a sound by shifting (up
or down) about a costant value of +x or -x [Hz] and then I want "Get resynthesis
PSOLA" or "LPC".
What is the best method to do it?

Thank you very much in advance.

Francesco Marzulli


[Non-text portions of this message have been removed]

I am doing a project within Praat in which I have recorded two people who
have trouble hearing the differences in syllable emphasis (they place
emphasis on the wrong syllable and then have a hard time correcting
it even when they hear it done correctly).

I have a recording of each person saying words that they have trouble
with, and within the recording, I have broken those words down into
syllables, and I selected each syllable to measure its duration and
intensity.

Is Intensity a combination of Duration and Amplitude?
Amplitude and Pitch?

Thanks!
LilAngelEyes247@...



[Non-text portions of this message have been removed]

Dear praat-users,

for a reason which I have not quite understood yet over the past 12
months a number of messages which I had approved for posting were not
sent the the list. The result is a backlog of 48 messages waiting for
approval.

In order not to swamp your mailboxes with so much mail I have decided to
send the messages out 10 at a time every three days over the next two
weeks - the first batch was just sent to you.

My sincere apologies to all those who posted messages which were never
sent to the list!

Daniel Hirst

--------------------------------------------------------------------------------\
-------------------
	      Equipe Prosodie et Représentation Formelle du Langage
	 CNRS Laboratoire Parole et Langage - Université de Provence
29 avenue Schuman     13621 Aix-en-Provence cedex 1        France
tel: +33-4-4295-3628 secr: +33-4-4295-3634 fax: +33-4-4259-5096
                                                 web: www.lpl.univ-aix.fr
--------------------------------------------------------------------------------\
-------------------
SProSIG list 				    www.lpl.univ-aix.fr/projects/sprosig
Speech Prosody 2002 April 11-13 	 www.lpl.univ-aix.fr/sp2002
--------------------------------------------------------------------------------\
-------------------

Hello list!!!
      Its nice to join this list for PRAAT users. I've just started using
such a versatile program. The most interesting feature I've found with the
program is signal display in many formats - spectrogram, wavelets,
cochleogram etc. I'll put forward the following questions for your best
advice -
1. Can we edit the wavelets using the program?
2. Could anyone please refer me to get more information about the
cochleogram?
Thanking you in anticipation,


M. Reddy Sivaprasad,
Junior Research Fellow,
All India Institute of Speech & Hearing,
Manasagangothri,
MYSORE-6     PIN: 570006
Karnataka    INDIA
Ph: +91-821-514618




_________________________________________________________________
MSN 8 helps eliminate e-mail viruses. Get 3 months FREE*.
http://join.msn.com/?page=features/virus&xAPID=42&PS=47575&PI=7324&DI=7474&SU=
http://www.hotmail.msn.com/cgi-bin/getmsg&HL=1216hotmailtaglines_virusprotection\
_3mf

Thanks. I'll try it.
mk

----- Original Message -----
From: "Hélène Loevenbruck" <loeven@...>
To: "Masahiko Komatsu" <koma2@...>; <praat-users@eGroups.com>
Sent: Monday, June 10, 2002 4:22 PM
Subject: Re: [praat-users] MOMEL algorithm


Here is a web page where you can find scripts that run LPL's MOMEL,
adapted to PRAAT by Guillaume Rolland:
http://www.icp.inpg.fr/~rolland/my_work/index.html
Cheers,
Hélène Loevenbruck

Masahiko Komatsu a écrit :
>
> Hello there,
>
> Does anyone know if a script to run the MOMEL algorithm by Praat (or maybe
> Matlab) is available? I would appreciate it if you offer me some
> information.
>
> Best,
> Masahiko Komatsu
> Sophia University/University of Alberta
>
> To Post a message, send it to:   praat-users@eGroups.com
> To Unsubscribe, send a blank message to:
praat-users-unsubscribe@eGroups.com
> To consult archives :            http://egroups.com/list/praat-users/
>
>
> Your use of Yahoo! Groups is subject to
http://uk.docs.yahoo.com/info/terms.html

Hi,

I've been trying unsuccessfully to find what the differences are between
weighting the average frequency with the power spectrum and with the absolute
spectrum, and what a power of 2/3 in the weighting corresponds to. Can
someone point me to a reference or have an intuitive explanation?

Thanks in advance,
Gina

I am working with spontaneous speech samples of varying length and would
like to calculate meaningful duration measures for voice and silence
periods.



As such, I would like to extract the data used to calculate the
voicelessness bar from the "to pitch (ac)" command, or in other words,
extract a frame by frame account of whether speech or silence is
present.



I am assuming I need to write a script to recreate this data, but I am
having trouble figuring out a command to make the silence/voice
determination in order to create the output.



Here is the simple script that I think does what I want, minus the
correct command.



count = Get number of frames



for i from 1 to count

             # Command to decide if frame i is silent or voiced

             printline # value 0 (silent) or 1 (voiced)

endfor





Any assistance would be greatly appreciated.  Also, if anyone has anyone
has any other suggestions for an appropriate duration measure for
spontaneous speech I would appreciate that input as well.



Thanks,

Michael



Michael.Schmidt2@...







[Non-text portions of this message have been removed]

I am working with spontaneous speech samples of varying length and
would like to calculate meaningful duration measures for voice and
silence periods.

As such, I would like to extract the data used to calculate the
voicelessness bar from the "to pitch (ac)" command, or in other
words, extract a frame by frame account of whether speech or silence
is present.

I am assuming I need to write a script to recreate this data, but I
am having trouble figuring out a command to make the silence/voice
determination in order to create the output.

Here is the simple script that I think does what I want, minus the
correct command.

count = Get number of frames

for i from 1 to count
   # Command to decide if frame i is silent or voiced
   printline # value 0 (silent) or 1 (voiced)
endfor

Any assistance would be greatly appreciated.  Also, if anyone has
anyone has any other suggestions for an appropriate duration measure
for spontaneous speech I would appreciate that input as well.

Thanks,
Michael

Michael.Schmidt2@...

Dear list,

Is there a specification for the PRAAT TextGrid file format somewhere out
there, i.e. somehting that tells me about how items, interval etc. have to
be encoded, what text encoding to use, where to use tabs and eol etc.?

Thanks for your help,
	 Thomas

---------------------------------------
Thomas Schmidt
SFB 538 'Mehrsprachigkeit' Teilprojekt Z
Tel: ++ 49 (040) 42838-6425
Fax: ++ 49 (040) 42838-6116
http://www.rrz.uni-hamburg.de/exmaralda
http://www.rrz.uni-hamburg.de/SFB538/
---------------------------------------

Greetings,

I can't get good quality greyscale output of Praat spectrograms to
print on my Laserjet 2100.

I am using the Praat picture window to export EPS files of
spectrograms on the highest quality setting.
The spectrograms look great in Praat's own picture window, but sending
the picture to the printer with the Praat print command produces an
image that is incorrectly dithered/overlapped in the horizontal
dimension, thus appearing with weird vertical strips of mismatched
greyscaling.
The greyscaling within each vertical strip appears fine grained.

Strangely, the EPS exported pictures look great on the screen, both in
Ghostview and also in the Xdvi LaTeX previewer after importing it into
a LaTeX document as a graphics object.
However, printing the LaTeX dvi file results in the image coming out
with exactly the same weird problem.

Is this some strange issue with my printer?

EPS files exported from PCQuirer (a rival speech software program, not
free) do not suffer from this problem, and they print beautifully.

Does anyone know what is going on, or have you experienced a similar
problem and can offer a suggestion?

I'd rather not go back to PCQuirer to remake all my spectrograms, I
was hoping to rely on Praat for my graphics.

-Sean Fulop

This is a followup to my previous post about bad spectrogram
printouts using my HP LaserJer 2100, a PostScript printer
(supposedly).

Here are the weird facts:
Praat spectrograms, converted to EPS files and incorporated into
PostScript documents, look great on the screen with PostScript
previewer software (i.e. Ghostscript) but look awful on the printed
page.
The images have vertical banded appearance, they are just hideous.
These results are obtained doing everything in Linux, which just uses
a straight printer dump to print PostScript files to any PostScript
printer, which mine is supposed to be.

Strangely, the same problems do not crop up with EPS spectrograms
produced using PCQuirer software, even when printed out in Linux.

Here is the fix, revealing something ungood:  When I take my
PostScript file with its incorporated Praat spectrograms over to
Windows and print it from there using GSview PostScript previewing
software (which also uses Ghostscript), THE SPECTROGRAMS COME OUT
BEAUTIFULLY, but the print options setting is crucial.
Here's the awful truth: you must select "Windows GDI Printer", which
uses Ghostscript to create bitmaps for each page, and prints these
using the standard Windows printer driver.  If you select "PostScript
printer", which does a straight dump like in Linux, the same old
problem resurfaces.

This proves that as usual the Hewlett Packard machine is not exactly
a PostScript printer, and these spectrograms only print correctly
when the HP special printer driver is invoked.  Of course, there is
no such thing for Linux, and the whole point of buying a PostScript
printer in the first place is so you don't need a printer driver.

It is still puzzling to me that only the Praat spectrograms cause
this printer problem to surface, while the PCQuirer spectrograms
don't seem to have the problem.

-Sean Fulop

I am working with spontaneous speech samples of varying length and
would like to calculate meaningful duration measures for voice and
silence periods.

As such, I would like to extract the data used to calculate the
voicelessness bar from the "to pitch (ac)" command, or in other
words, extract a frame by frame account of whether speech or silence
is present.

I am assuming I need to write a script to recreate this data, but I
am having trouble figuring out a command to make the silence/voice
determination in order to create the output.

Here is the simple script that I think does what I want, minus the
correct command.

count = Get number of frames
for i from 1 to count
     # Command to decide if frame i is silent or voiced
     printline # value 0 (silent) or 1 (voiced)
endfor

Any assistance would be greatly appreciated.  Also, if anyone has
anyone has any other suggestions for an appropriate duration measure
for spontaneous speech I would appreciate that input as well.

Thanks,
Michael
Michael.Schmidt2@...

Anyone know whether one can write into a Praat script a call to a
procedure that lives in another script?