Praat

For the discussion of language mechanics, grammar, vocabulary, trends, and other such linguistic topics, in english and other languages.

Moderators: gmalivuk, Moderators General, Prelates

fənɑlədʒɪst
Posts: 64
Joined: Sun Aug 01, 2010 12:14 am UTC

Praat

Postby fənɑlədʒɪst » Sun Dec 05, 2010 9:00 pm UTC

Hello all.

A few people have mentioned that they'd be interested in learning how to use Praat, a program often used for phonetic analysis of speech. If there's already a topic on this, someone please lock the thread or something like that :P Otherwise, I'll do my best to help out people here.

I don't have the time to write out everything that I know at the moment, but I can start by telling people how to record a sound file using Praat. Or, alternatively, how to open a .wav file to analyze.

Recording a .wav file

You must have a microphone or some sound input device plugged in, or Praat will complain at you.

1. Open Praat.
2. On the "Praat Objects" window, click the "New" option at the top.
3. From the list, choose "Record mono sound..."
4. If the defaults are fine with you (I've never had to change them), click the "Record" button in the bottom left corner of the "SoundRecorder" window.
5. Say some words. I suggest, if this is your first time looking at spectrograms, to keep your words isolated. It's difficult to see word boundaries if you say entire sentences. Some would say there's no such thing as a word boundary in a spectrogram, heh.
6. When you've finished saying the words you'd like to see in Praat, click "Stop."
7. Click "Play" to make sure your recording worked the way you wanted and you can clearly hear your words.
8. If you're satisfied, you can name your recording by changing the "untitled" in the "Name:" area at the bottom right. Then, click "Save to list & Close."

Your "SoundRecorder" window should have closed, leaving you with a new object in your "Praat Objects" window.

9. If your object isn't already selected (blue), select it by clicking it.
10. Click "Edit" from the menu on the right of the "Praat Objects" window. It is the second option from the top.

You should now be able to see your recording in a new window called "Sound <Name of your file>". The bottom half of this window is the really interesting part - the spectrogram. This is where you'll do all sorts of stuff, like look at pitch, formants, and duration of consonant or vowel segments.

Loading a .wav file into Praat

1. Open Praat
2. On the "Praat Objects" window, click the "Read" option at the top.
3. Click "Read from file..."
4. Navigate your computer and find the .wav file you would like to open. Click open, as you would for any other program :)

You should now have a new object in your "Praat Objects" window titled something like "Sound <Name of your file>". If you would like to open this file to view it in a spectrogram, look at steps #9 and #10 above.

Saving your Praat recording to a .wav file

This is assuming you already have your recording in your "Praat Objects" window.

1. If your object isn't already selected (blue), select it by clicking it.
2. On the "Praat Objects" window, click the "Write" option at the top.
3. Choose the "Write to WAV file..." option from the menu.
4. Name your file, choose the directory to which you'd like to save it, and click "save".

You may reopen your .wav file at a later date by following the directions for loading a .wav file into Praat, shown above.

Looking at sounds in your Praat Recording!

Congratulations. You now get to check out some sounds in your spectrogram! For the purposes of this section, we'll be examining [b] and [β] in Standard Japanese.

This is assuming that you've made a Praat recording and it is currently in your Praat Objects window.

1. If your object isn't already selected (blue), select it by clicking it.
2. Click "Edit" to the right in the Praat Objects window. It should be the second option from the top.
3. Select a part of your recording in the Sound .... window by clicking somewhere in the spectrogram and dragging your cursor to the other end of what you'd like to see. Remember, the spectrogram can only display segment analysis of 10 seconds or less.
4. Click "sel" for 'Selection' in the bottom left corner to zoom in to your selection.

Hopefully, you can now see the word or words you recorded in your Praat recording in the analysis (bottom) half of your Sound .... window. Basically, it should look like a lot of black and grey lines and stuff, much like those shown in the wikipedia article for "spectrogram" that I linked above. I'll now take just a short time to explain one thing you can tell from spectrograms- whether a sound is a stop or a fricative. Refer to the picture below.

Image

First of all, the letters and separations you see in this picture are called a Textgrid, and I'll explain how to make one later. For now, we'll just use the Textgrid to call your attention to the [β] in the first spectrogram of [noβot:a] 'climbed' and the [b] in the second spectrogram of [beŋkʲo:] or [beŋkjo:] 'study'. Generally, linguists analyze both of these sounds in Standard Japanese as allophones of the phoneme /b/, as they do not distinguish words and [nobot:a]/[noβot:a] both occur in speech, although [β] rarely occurs word initially, of course (for more information, read up on intervocalic spirantization).

The difference between these two sounds in the spectrogram is that [b] is mostly white while [β] has lots of black, which represents friction. Logically, this makes sense, as [b] is a stop consonant (stoping airflow) and [β] is a fricative (allowing relatively free airflow). This examination can also show you the differences between velar stops/fricatives, as well as others. Affricates, as you can guess, show up as a short stop followed by a fricative release. I'll see if I can post a picture of Japanese's alveolar affricate [ts] later.



I hope you guys have found this first step into Praat useful. Admitted, you can't do much yet in terms of phonetic analysis, but knowing how to use Praat first is really helpful. In a bit, I'll work on sections on how to use the spectrogram to actually tell things about the sounds you're looking at, assuming this topic stays up. Let me know if you guys have any questions or see something I've missed.

-fənɑlədʒɪst

fənɑlədʒɪst
Posts: 64
Joined: Sun Aug 01, 2010 12:14 am UTC

Re: Praat

Postby fənɑlədʒɪst » Fri Dec 17, 2010 2:51 pm UTC

Added another section to the Praat How-To thingy.

posentin
Posts: 8
Joined: Sun Aug 22, 2010 2:03 am UTC

Re: Praat

Postby posentin » Sat Dec 18, 2010 3:18 am UTC

I always wondered how one would read consonants from a spectrogram.

fənɑlədʒɪst
Posts: 64
Joined: Sun Aug 01, 2010 12:14 am UTC

Re: Praat

Postby fənɑlədʒɪst » Sat Dec 18, 2010 10:22 am UTC

There are a few other signs to look at too, like velar pinches for velar stops and rising formants for labials... really high frequency friction for sibilants, but I can't help but feel that spectrograms are lacking in a lot of ways when distinguishing consonants. Basically, you need to have a good ear for phonetics too so you can listen to the recording. Other than velars and labials and sibilants, there aren't really that many defining characteristics between consonants that I can point out- such as the difference between a voiced uvular fricative and a voiced postalveolar fricative, for example. It's pretty hard to tell the difference between various sibilants too, although there are some things, like how a postalveolar sibilant is a lower frequency than an alveolar sibilant, and labiodental sibilants are even lower frequency.

Telling the difference between postalveolar and alveolopalatal fricatives is basically impossible in spectrograms, and the only thing I can suggest is learning Japanese, Korean, or Mandarin and trying to distinguish the [ʃ] and [ɕ] sounds. [m] and [n] in isolation are also basically the same. The main ways of distinguishing [m] and [n] in speech are from the vowel formant transitions before and after the nasal consonant.

Neener
Posts: 1
Joined: Mon Feb 28, 2011 1:34 pm UTC

Re: Praat

Postby Neener » Mon Feb 28, 2011 1:37 pm UTC

Hi,
I am wondering how I can tell the difference between the voiced/voiceless and lenis/fortis fricatives /z/, /z_0/ and /s/.
Thanks,
Neener

kirkedal
Posts: 22
Joined: Fri Jun 26, 2009 1:52 pm UTC

Re: Praat

Postby kirkedal » Wed Apr 20, 2011 11:08 pm UTC

A nice exercise is to record a sentence spoken very quickly and then reducing the the speed of the recording.

User avatar
Microscopic cog
in his catastrophic plan
Posts: 769
Joined: Thu Oct 07, 2010 12:09 pm UTC

Re: Praat

Postby Microscopic cog » Sat Mar 24, 2012 5:14 pm UTC

Hey just necro-ing to share a funny encounter at the University of Amsterdam.

I was attending the orientation day for Language Science, and there were a couple of teachers talking about the course, and what they did there. One of them came from a physics education but somehow ended up working with speech recognition and so on at the University, which was really interesting to hear. Anyway, somewhere along the way they mention that he developed some software, turns out later the person who was speaking was David Weenink (pic), one of the developers of PRAAT. I told him about how I thought that was pretty cool, and said I heard of it via XKCD, which sadly, he had not heard of. Oh well, it was fun.

Image
Spoiler:
Interviewer: Some people say they can’t understand your writing even after they read it two or three times. What approach would you suggest for them?

William Faulkner: Read it four times.

User avatar
cattfiiish
Posts: 3
Joined: Thu Jan 23, 2014 7:46 pm UTC

Re: Praat

Postby cattfiiish » Thu Jan 23, 2014 7:57 pm UTC

Hi!!

I am using praat to analyze animal sounds. Currently I am interested in measuring formants (I am using "To: Formants (burg)"). In addition to the formant frequencies, I would like to get an estimate of the relative "amplitude" or "intensity" contribution of each formant. I see that there is an intensity measure in my "list" for the burg formants but I do not understand what it is telling me. It gives me one intensity measure per "window" ... and the intensity listing are on the order of "0.00020" which doesn't match what I would expect.

I suspect some frequencies are being emphasized more than others (particularly formants between 1000hz and 4000hz (my animals have a very high voice)) and would like to quantify this using praat.

Anyone have any suggestions? Am I asking something of praat that is not possible?

Thank you humans!

C

User avatar
cattfiiish
Posts: 3
Joined: Thu Jan 23, 2014 7:46 pm UTC

Re: Praat

Postby cattfiiish » Sat Jan 25, 2014 9:20 pm UTC

Oh! One more question.

Is there a way to calculate area under the LPC curve? Where I can I get the equation for the curve? If I had that equation I could just take the integral between frequencies...

Cora

User avatar
Jplus
Posts: 1721
Joined: Wed Apr 21, 2010 12:29 pm UTC
Location: Netherlands

Re: Praat

Postby Jplus » Mon Jan 27, 2014 12:38 pm UTC

Oh, the nostalgia. I studied linguistics from 2005-2009 and I used Praat quite a few times. Fond memories.

(cattfiiish: sorry for not answering your question, right now I'm rusty and it would take me a lot of time to figure it out.)
"There are only two hard problems in computer science: cache coherence, naming things, and off-by-one errors." (Phil Karlton and Leon Bambrick)

coding and xkcd combined

(Julian/Julian's)

User avatar
cattfiiish
Posts: 3
Joined: Thu Jan 23, 2014 7:46 pm UTC

Re: Praat

Postby cattfiiish » Mon Jan 27, 2014 9:09 pm UTC

It is ok. I think I have a way to get relative intensity just using the spectral slice. So Burg's mystery intensity isn't needed for the analysis but ...

I am curious, though, about what the hell the intensity measure is. I obtained it by using

Objects -> Formants & LPC -> To Formants (burg)
Then back in Objects there is a "formant object" and I highlight it
Then go:

Objects -> List -> and I make sure I click "include intensity" and I get the image below.

What does that intensity measure mean? I have one per "analysis window" ... Help?
Attachments
mysteryintensity.png

littlemorte
Posts: 1
Joined: Thu Sep 04, 2014 5:47 am UTC

Re: Praat

Postby littlemorte » Thu Sep 04, 2014 5:51 am UTC

Funny story... PRAAT doesn't want to open my WAV files.

Anyone have an idea how I can fix this?

Thanks!

PhantomSoul
Posts: 2
Joined: Sat Mar 19, 2011 3:36 am UTC

Re: Praat

Postby PhantomSoul » Mon Sep 08, 2014 8:08 pm UTC

...Can you give any more details? Opening a very long sound file with the "Read from file..." option or use ctrl+O (or Mac equivalent) won't work, you need to use "Open long sound file..." (ctrl+L / Mac equivalent). Without knowing more about this case, it's hard to give more help! :S

User avatar
Logological
Posts: 9
Joined: Tue Mar 17, 2015 3:52 pm UTC

Re: Praat

Postby Logological » Sat Mar 28, 2015 8:23 pm UTC

Does anyone here know how I can colourize the spectrograms produced by Praat? Bartlomiej Plichta used to provide a script for this (which shells out to ImageMagick) but unfortunately his website has been down for months and there doesn't seem to be any way of contacting him.


Return to “Language/Linguistics”

Who is online

Users browsing this forum: Google [Bot] and 7 guests