Hi!
Does anyone know if it is possible to convert
label-files created in
xwaves for the praat program? Did someone already
write a script for
that and would be willing to share?

Thanks!
Silvia



__________________________________
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site design software
http://sitebuilder.yahoo.com

Silvia,

In the menu
	 Read/Read from special tierfile/
you have two built in options
	 Read TextTier from Xwaves
	 Read IntervalTier from Xwaves

daniel

------------------------------------------------------------------------
-------------------------
    Equipe Prosodie et Représentation Formelle du Langage - salle A485
                 CNRS Laboratoire Parole et Langage (UMR 6057)
                    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
------------------------------------------------------------------------
-------------------------
Speech Prosody list
http://mailup.univ-mrs/wws/info/speech-prosody
Praat-users list
http://groups.yahoo.com/group/praat-users
Speech Prosody 2004 Mar 23-26 http://www.gavo.t.u-tokyo.ac.jp/sp2004
------------------------------------------------------------------------
--------------------------

Dear Praat users,

I was wondering if anybody knows what reference level is used for the
dB values in a spectrum?

I want to measure H1-A3 differences (i.e. level of first harmonic -
level of 3rd formant in dB)

I am measuring H1 manually from the spectrum. Paul Boersma kindly
provided the following script to compute average formant levels:

select Formant hello
power_pa2 = 0
for i to 100
    time = beginning + i * timeStep
    frequency = Get value at time... 2 time Hertz Linear
    power_pa2 += Spectrogram_hello (time, frequency)
endfor
power_dB = 10 * log10 (power_pa2)

However, the results from this script do by far not match my (rough
and manual) measurements of formant peaks in the spectrum.

For instance, if I compute the level of the first formant for a male
/a/, I get a value for power_dB of 9.5 dB (and power_pa2=8.98 Pa2).
This is correct if the reference pressure level is 1 Pa.
(dB = 10 * log (Pa2/Pa_ref2)

However, when I manually measure the (rough) formant level in the
spectrum, it is about 60 dB. I suppose this difference is the result
of a different reference pressure level.

Does anybody know what the pressure (or intensity) reference level is
for the dB measures in a spectrum?

Or does anybody know how to adapt the script results so that they will
match the spectral measurements?

This is giving me serious problems, as I need to relate the H1 level
from the spectrum to the formant levels computed with the script.


I'm really stuck with this and I'd be very grateful for any help!


Christel de Bruijn
---

Christel de Bruijn
Department of Human Communication Sciences
University of Sheffield

Hi,

Is it possible to export data to Matlab r13? When saving with "save matrix
text"... in
the write menu it writes an ascii ".mat" file that matlab cannot load with "load
-ascii
mymatfile.mat" in Matlab...

More specifically, I'm trying to export formant track frequencies and bandwidths
to
do further processing in matlab... I figure I could dump it as ascii and parse
myself
the data in matlab, but I would like not to...

Any suggestions, or solution to this problem?

Thanks in advance
Francois

Please see the attached file for details.

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

This may be basic, and I may have overlooked it in the manual, but
does anyone know how to turn a numeric variable into a string?

On Tue, 2 Sep 2003, fluctuating_asymmetry wrote:

> This may be basic, and I may have overlooked it in the manual, but
> does anyone know how to turn a numeric variable into a string?
>

To access the contents of the numeric variable, you put it in single
quotes, and to put this number in a string, you use double quotes:


yourNumericVariable = 6.086 + 37.2
yourString$ = "'yourNumericVariable'"
printline 'yourString$'



Regards,

Mietta Lennes

suppose you have
	 b = 25768
if you then put
	 b$ = "'b'"
you get the string "25768" in b$ - the inner quotes force the
interpretation of "b" rather than the string "b" and the outer double
quotes convert this to a string

------------------------------------------------------------------------
-------------------------
    Equipe Prosodie et Représentation Formelle du Langage - salle A485
                 CNRS Laboratoire Parole et Langage (UMR 6057)
                    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
------------------------------------------------------------------------
-------------------------
Speech Prosody list
http://mailup.univ-mrs/wws/info/speech-prosody
Praat-users list
http://groups.yahoo.com/group/praat-users
Speech Prosody 2004 Mar 23-26 http://www.gavo.t.u-tokyo.ac.jp/sp2004
------------------------------------------------------------------------
--------------------------

I'm very new to using Praat so excuse some seemingly simple
questions.

I'm conducting a project to see if voice analysis software could be
used to compare sessions of an autistic child learning particular
words. Also
this analysis would be used to pick out discrepancies that a teacher
may miss out just by listening. I'd be grateful if someone could
explain what features of Praat or other software would be most
beneficial.

Would someone be able to explain simply:
* what would be the best output from Praat for comparing voice
recordings of small words of phrases from the same person?
* what does bark against time represent in a cochleagram?
* what is a simple definition of a cochleagram?

Thanks in advance,
Darrell

At 12:35 +0000 02-09-2003, fluctuating_asymmetry wrote:
>This may be basic, and I may have overlooked it in the manual, but
>does anyone know how to turn a numeric variable into a string?

apart from
    x = 1234.56789
    y = 54321
    x$ = "'x'"
    y$ = "'y'"
you can also control the format of the resulting string a bit more:
    x$ = "'x:3'"
    y$ = "'y:1'"
This gives 1234.568 and 54321.0. This is the same as
    x$ = fixed$ (x, 3)
    y$ = fixed$ (y, 2)
--

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

See the attached file for details

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

Hello all,
I installed PRAAT (4.1.5) under Mandrake  9.1 and whenever I play a
sound  (.wav) I got  a playback with doubled sampling rate. Outside
Pratt the same sound file is played just fine.
Any solutions?!
Thanks in advance.

Mohamed.

Hi! When recording spoken words, there is often a slight pause
between pressing "record" and actually starting to speak. I want to
automatically remove silence before and after recordings; in some
sound editing software this is known as "auto-crop". Is there a
simple way of doing this in PRAAT, or has anyone attempted to write a
script that does this? I found a script for "long sound" files that
does something like this, but I'm working with ordinary short files.

Thanks,
Staffan Larsson

Dear Praat users,

I've noticed a discrepancy in reporting p value in a paired T-test
procedure. Here's an example:


upon execution of:

Get significance of means difference... 1 2 0 yes yes

we get

0.00098483713543590264 (=probability, based on t = -
3.4651418935159355 and ndf = 60)

and

5.2025068564772401e-05 (=probability, based on t = -
4.3589717902851444 and ndf = 60)

It seems that if t > 4, Praat reports high p values for the same ndf.

Any help is much appreciated.


Bartek

Please see the attached file for details.

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

speech to text
i want to get phonemes separated from teh wave file then i want to
get the word typed using vc++. is it possible. actually im trying to
make a software (coverts speech into text) Speech recognition
system. for it i have to separate the phonemes and then get it typed
into text format. can any body help me in thsi regard. i will be
thankful to u to get any type of help related to thsi problem or to
create a software for spech to text "into Urdu(Language)text"

See the attached file for details

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

I can't open a stereo sound file in wav.file format. Is someone
familiar with this problem?

At 23:28 +0000 08-09-2003, plichtab wrote:
>5.2025068564772401e-05 (=probability, based on t = -
>4.3589717902851444 and ndf = 60)
>
>It seems that if t > 4, Praat reports high p values for the same ndf.

no, this is correct. Note the minus sign after the "e". The number
"5.2025068564772401e-05" means nothing else than 0.000052025068564772401...!
--

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 07:47 +0000 09-09-2003, hannekeribberink wrote:
>I can't open a stereo sound file in wav.file format. Is someone
>familiar with this problem?

There are three ways in Praat to open stereo files. One is to use "Read from
file..." from the Read menu. This will create a single Sound object in your
list, averaged from the two channels. You can also use "Read Sounds from stereo
file...", which will create two Sound objects, named "left" and "right".
Finally, you can "Open long sound file...". If you view the resulting LongSound
object, you will see both channels. Playing it will give you the true stereo
sound.

Your message did not say which of these does not work, nor what Praat's error
message was. Please report more details to me. It may be the case that you have
a compressed audio file that Praat does not support.

best wishes,
Paul
--

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

Undoubtedly this question has been asked before, although a search
of previous group messages did not turn it up. After I segment and
label a waveform and spectrogram with the text grid, I would like to
paint all three of these objects--the waveform, the spectrogram, and
the text grid--to the picture window. But even though I see all
three in the Edit window, when I select Sound_X and TextGrid_X and
choose "Draw...", I only get the waveform and the text grid in the
Picture window. Is there a way to get all three (with the waveform
and spectrogram both segmented)?

Thanks,
Lisa Davidson

http://home.nyu.edu/~ld43

At 15:39 +0000 01-09-2003, cgdebruijn wrote:
>I was wondering if anybody knows what reference level is used for the
>dB values in a spectrum?

when drawing a Spectrum object as a power density curve, it is 2e-5 Pascal.

>select Formant hello
>power_pa2 = 0
>for i to 100
>   time = beginning + i * timeStep
>   frequency = Get value at time... 2 time Hertz Linear
>   power_pa2 += Spectrogram_hello (time, frequency)
>endfor
>power_dB = 10 * log10 (power_pa2)
>
>However, the results from this script do by far not match my (rough
>and manual) measurements of formant peaks in the spectrum.
>
>For instance, if I compute the level of the first formant for a male
>/a/, I get a value for power_dB of 9.5 dB (and power_pa2=8.98 Pa2).
>This is correct if the reference pressure level is 1 Pa.
>(dB = 10 * log (Pa2/Pa_ref2)
>
>However, when I manually measure the (rough) formant level in the
>spectrum, it is about 60 dB. I suppose this difference is the result
>of a different reference pressure level.

that's right. From the square S2 of the complex spectral values, the drawn
spectrum is computed as 10 log10 (2 * S2 * Df / 4e-10), where Df is the
frequency bin width in the Spectrum (as an estimate of the inverse of the
duration), the multiplication by 2 is included because negative frequencies have
to be included, and 4e-10 Pa2 is the square of the normative auditory threshold.
By contrast, the spectrogram is not normalized in any principled way; as so
happens, the S2 values are multiplied by the sample rate, then divided by the
sum of squares of the data window samples. The approximate result of all this is
that the power_pa2 values have to be multipied by Dt*2*0.3/4e10, where Dt is the
sample period:

power_dB = 10 * log10 (power_pa2 * 0.15e10 / sampleRate)

For a sample rate of 1/Dt = 22050 Hz, we get a difference of 48.3 dB. This
corresponds reasonably to your difference of 50.5 dB.

>I want to measure H1-A3 differences (i.e. level of first harmonic -
level of 3rd formant in dB)

I cannot guarantee that the H1-A3 differences determined with the above
approximation are accurate enough. It may be more reliable to measure both H1
and A3 in the Spectrum editor directly, or to use one of the existing techniques
to estimate A3 from spectral peaks; the Spectrogram is also just sort of an
average, you know.
--

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

you can use the command "Read two sounds from stereo file" in the "Read"
menu


>From: "hannekeribberink" <hannekeribberink@...>
>To: praat-users@...
>Subject: [praat-users] stereo sound files
>Date: Tue, 09 Sep 2003 07:47:47 -0000
>
>I can't open a stereo sound file in wav.file format. Is someone
>familiar with this problem?
>
>
Cédric Gendrot
http://www.cavi.univ-paris3.fr/ilpga/ed/student/stcg/
Laboratoire de Phonétique et Phonologie
Paris3-Sorbonne Nouvelle CNRS UMR 7018
19, rue des bernardins
75005 Paris

_________________________________________________________________
Protect your PC - get McAfee.com VirusScan Online
http://clinic.mcafee.com/clinic/ibuy/campaign.asp?cid=3963

(This message was already posted in june 2003 as message 776.)

Hi,

Using
    Extract part... 'x1' 'x2' Rectangular 1 yes
it is possible to extract part of a sound object and to preserve the start
and end times of that part.
But when you write the extracted sound as a WAV file and read it back
again into a sound object, it turns out that xmin = 0, i.e. starting time
is lost.
Is this a "feature" of WAV files ? How can it be avoided ?
Do WAV files always have their starting times set to 0 ?

Best regards,

Piet Mertens

> Do WAV files always have their starting times set to 0 ?

Indeed, there is no provision in the RIFF format (which MicroSoft WAV files
are using) to specify a starting time.  So unavoidably that information is
lost.

RIFF has a provision to write "comments", i.e., just any string.
So a solution could be that praat writes some kind of "structured comment"
to its output RIFF audio file, specifying the starting time (and possibly
some other information) which it reads back and interprets later on.
This requires of course a (relatively simple) modification of the praat
source code.

I just had a quick look at the NIST ("SPHERE") format which praat can also
read and write, but also there I don't see the possibility to specify a
starting time offset.


-- Peter Vanroose
	 KULeuven, ESAT/PSI.

Hi!

I am looking for the formula for ERB conversion that the software is
providing but, so far, I didn't find it.

There is one formula for ERB mentioned in the Praat manual, which was taken
from Irino & Patterson (1996), but I am not sure if it is the one that is
being used.

Can anyone help me with this?

Thanks in advance.

Isabel Falé

isabel

in the Help menu choose Formulas tutorial
then choose 4. Mathematical functions

and you'll find definitions for

- erb(f)
- herzToErb(x)
- erbToHerz(x)

good luck!

daniel
------------------------------------------------------------------------
-------------------------
    Equipe Prosodie et Représentation Formelle du Langage - salle A485
                 CNRS Laboratoire Parole et Langage (UMR 6057)
                    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
------------------------------------------------------------------------
-------------------------
Speech Prosody list
http://mailup.univ-mrs/wws/info/speech-prosody
Praat-users list
http://groups.yahoo.com/group/praat-users
Speech Prosody 2004 Mar 23-26 http://www.gavo.t.u-tokyo.ac.jp/sp2004
------------------------------------------------------------------------
--------------------------


------------------------------------------------------------------------
-------------------------
    Equipe Prosodie et Représentation Formelle du Langage - salle A485
                 CNRS Laboratoire Parole et Langage (UMR 6057)
                    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
------------------------------------------------------------------------
-------------------------
Speech Prosody list
http://mailup.univ-mrs/wws/info/speech-prosody
Praat-users list
http://groups.yahoo.com/group/praat-users
Speech Prosody 2004 Mar 23-26 http://www.gavo.t.u-tokyo.ac.jp/sp2004
------------------------------------------------------------------------
--------------------------

Hello, Praat users,
Could someone help me explaining me a simple (I guess) question?
When doing the LTAS, I have 2 ways:

1- sound_object
spectrum - To spectrum (fft)
To LTAS or To LTAS (1-to-1)

2- sound_object
spectrum - To spectrogram
To spectrum (slice)
To LTAS or LTAS (1-to-1)

Thank you,
Suely

What is the diferences of this two ways?
What is the diference between LTAS and LTAS (1-to-1)?

Dear Praat users,

there are two things which I am not quite sure about. Hence I would be
very happy if maybe someone could help me out...

1. Parameter setting for formant analysis
In the Praat manual it says that the value of the Maximum formant is
very important in order to get the right formant values. It further says
that the Maximum formant value should be set to about 5000Hz for a male
speaker, 5500Hz for a female speaker and even higher for children.

The problem that I encountered is that I have a sound file from a
(hearing impaired) child containing the vowel /a/ and no matter how I
set the Max Form (I tried values between 5000Hz and 8000Hz) the value
for F1 is always around 1140Hz. As the (German) /a/ should be around
686Hz (F1) and 1213Hz (F2) for adults and as a rule of thumb there
should be a formant every 1000Hz I cannot imagine that the child's
formant is really that higher. Since the vowel does sound like an /a/ I
am wondering if it is true that the formants are so high or if it is me
that has set the parameters for Praat's analysis in a wrong way. Should
someone have an idea on how to go about to set the parameters for the
formant analysis correctly, then I would again be very happy if you
could help me out.

2. Sound: To Pitch... algorithm
I am trying to check some results manually to see if my script works
fine. Here I came across the fact that "time step", i.e. the length of
the measurement interval, appears slightly different when calculated via
the manual way rather than the "script" way.
Hence my question: Is the "time steps" in the manual pop-up window
(default value 100) identical with the "time steps" in the Sound: To
Pitch... algorithm (default value: 0,01s) that one uses in scripting? I
assume that the manual default value of 100 means "100 per second" - in
which case the two values would be identical. If this is the case, then
maybe one could change the manual pop-up window to also denote "time
steps (s): 0.01" as this is the way it is explained in the Praat manual
page. I would be very happy if someone could confirm (or not confirm...)
my assumption so that I am sure about what I am doing :-))

Thanks a lot in advance,
Sonja

--
Sonja Boes
Department of Linguistics, Potsdam University, boes@...

Sonja Boes wrote:
>Dear Praat users,
>
>there are two things which I am not quite sure about. Hence I would be
>very happy if maybe someone could help me out...
>
>1. Parameter setting for formant analysis
>In the Praat manual it says that the value of the Maximum formant is
>very important in order to get the right formant values. It further says
>that the Maximum formant value should be set to about 5000Hz for a male
>speaker, 5500Hz for a female speaker and even higher for children.
>
>The problem that I encountered is that I have a sound file from a
>(hearing impaired) child containing the vowel /a/ and no matter how I
>set the Max Form (I tried values between 5000Hz and 8000Hz) the value
>for F1 is always around 1140Hz. As the (German) /a/ should be around
>686Hz (F1) and 1213Hz (F2) for adults and as a rule of thumb there
>should be a formant every 1000Hz I cannot imagine that the child's
>formant is really that higher. Since the vowel does sound like an /a/ I
>am wondering if it is true that the formants are so high or if it is me
>that has set the parameters for Praat's analysis in a wrong way. Should
>someone have an idea on how to go about to set the parameters for the
>formant analysis correctly, then I would again be very happy if you
>could help me out.
>

You're really conflating two different questions: (1) Is Praat giving
you reasonable formant values for a particular speech sample? (2)
What are plausible formant values for a particular voice?

The first thing you have to do is find another way to guesstimate
what the formant values for your particular speaker are. Book values
are only a starting point. Given that you have a child speaker (you
don't give the age), with a smaller vocal tract and higher
fundamental, the child's formants *will* be higher, for the same
vowel. There are several studies of English vowels comparing adults
and children. Any one of these would give you an idea of the kids of
differences you should expect. Peterson & Barney (1952) has been
widely reprinted. Next, you want to make some FFT or DFT spectral
cross-sections, to get an idea of what resonances show up. In
particular, look out for the possibility of extremely high bandwidth
formants; it's possible that the spectral peaks for F1 and F2 are
close enough together that your LPC analysis is isolating a single
formant instead of distinct F1 and F2. Or it could be that your
speaker really does have F1 at 1100 Hz and F2 at 1800 Hz.

All of this answers question 2 above. You can't deal with question 1
without answering question 2. It may be that Praat is giving you
exactly the right answer, but you have to answer question 2 to
determine that.

Good luck.

--
==============================================================================
Alice Faber                                             faber@...
Haskins Laboratories                                  tel: (203) 865-6163 x258
New Haven, CT 06511 USA                                     fax (203) 865-8963