TUTORIAL: Analyzing pitch with Praat

[Jennygirl]

A lot of people have been expressing interest in using Praat to analyze vocal pitch and resonance, so I though it might be helpful to do a little tutorial... Maybe this will help with getting through some of that initial confusion

As a disclaimer- I have limited knowledge of this program, so please share any tips or shortcuts to these steps or other helpful features / methods. I am still learning, too! Anyway, here we go! Also, I am doing this on a Mac... so, apologies if it is slightly different for Windows.



How to use Praat to find your fundamental frequency:

1. Download and install Praat (these links may become dead, so just google search "download praat mac" or windows)
mac:
http://www.fon.hum.uva.nl/praat/download_mac.html
win:
http://www.fon.hum.uva.nl/praat/download_win.html

2. Open it, and locate the "Praat Objects" window

3. Click the menu "New" at the top left, then select "Record mono Sound" (this is Apple+R on a mac, I'm guessing for "record")



4. Hit "Record", then record your voice, then click "Stop"



5. Save the recording to the Praat Objects window by hitting "Save to list & Close" at the bottom right

6. Re-locate the "Praat Objects" window and make sure the last item in the list is selected



7. Click "View & Edit". This will open your recording into the window where you can see your pitch on a blue graph below.



8. In the new window click the "View" menu, then select "Show Analyses"



9. This is where you can tell Praat what and what not to show. I have mine showing "Show Spectrogram" and "Show Pitch". Also, set "longest analysis" to 300.



10. Click ok and locate the blue graph in the window you were in previously. That is your pitch.

11. Click and drag over the areas on the blue graph to analyze either words or the whole thing if you want. The average pitch for the selected section appears on the right.



Resonance analysis: coming soon!



I hope this helps! Let me know how it works or doesn't work for you and I'll try to keep this tut updated

Happy voice testing, and may all of your voices sound undeniably identified!

[sarahb]

I tried using Praat before and found it very confusing. This has totally helped! Now I can get more accurate measurements of my frequency. So very helpful, thank you

[Robin Mack]

You did it!!!  Thank you, o thank you!

*happy dance*

=)

[Adam (birkin)]

oooh, this looks like something I want to try out.

[sarahb]

One thing I've noticed is that here and there it will put the blue line way high, like in the 4000+Hz range. I'm assuming the blue line tracks the average frequency at any given point. I'm wondering if you've seen that as well Jenny? When I highlight an area with one of those skewed portions it seems to skew the average frequency (increasing it, obviously). When I just highlight areas around it the results seem to average out better.

I did notice that at least two of the times it seemed to be right where an "S" sound was, so I'm thinking that it was catching the high pitch of the "S". However, that wasn't the case in all of the areas. In any case, I'm having a blast using this program now.

[Carrie Liz]

I'm still really confused about the resonance thing.

When I click "show spectral power," it gives me a scientific-notation number like 6.123692172698402e-006 Pa2/Hz, then says "at time = 11 seconds and frequency = 3500 Hz"

I assumed that this Hz number was what you were taking about, but upon further analysis, the Hz number changes completely independent of the graph itself, and is just a representation of where on the vertical axis my cursor is. Like, if my cursor point is near the top of the screen, regardless of what the graph says, the number there would be 5000 Hz. And if it's at the very bottom, again completely independent of the graph itself, it would say 0 Hz.

So, um, am I doing something wrong?

[sarahb]

I'm still really confused about the resonance thing.

When I click "show spectral power," it gives me a scientific-notation number like 6.123692172698402e-006 Pa2/Hz, then says "at time = 11 seconds and frequency = 3500 Hz"

I assumed that this Hz number was what you were taking about, but upon further analysis, the Hz number changes completely independent of the graph itself, and is just a representation of where on the vertical axis my cursor is. Like, if my cursor point is near the top of the screen, regardless of what the graph says, the number there would be 5000 Hz. And if it's at the very bottom, again completely independent of the graph itself, it would say 0 Hz.

So, um, am I doing something wrong?

I see the same thing.

[Ltl89]

Thanks for the tutorial Jenny.  I've been working on my voice, but I've yet to record myself.  I'd hate to hear what the playback sounds like.  Pus, I do it in my car (for the most part). However, I suppose it's part of the process, so I should give it a try.  Looks like a good and confusing program, lol.  Do you know if it runs on linux?  And if not, is it configurable with Wine? 

[anjaq]

Thanks a lot! Can you explain the part on resonance a bit more? I tried that but it seems the numbers do not only depend on the timepoint (x-axis) but also on where you click in the spectrum on the y-axis. So iWhere do you have to click to get these numbers? So if I click on the blue line, it is like 500 or so. By the way in my settings - I dont know if I accidentially changed something, that spectral power number appears as a red number on the left if you click in the spectrum.

To zoom in fast by the way you can mark with the mouse and then press CTL-N to get the part of the recording you want to analyze.

[KatelynRain]

Wow!  I didn't even know this program existed!! My speech therapist mentions that I could stop his speech workshop at Ithaca college to have my voice measured, but with this I could check for myself!!  So cool! Thanks for posting!!!!

[Aina]

This is just what I needed Jenny! Your are awesome girl!

So my pitch seems to be around 240-260 hz, which Is good.

But the resonance part is the real nice part, it seems I am just below the female mark with an average 2600-2700 guess that is where I need to improve!

Then again I am just about getting over a cold...got overly excited seeing this tutorial going have to try again when I am fully recovered.

-hugs- 

[Jennygirl]

I am looking into the resonance analysis. The way I was doing it is definitely not proper!

I've been looking into it and I think I've found it... You can do it one of two ways: either by taking a spectral slice and seeing where the harmonics are strongest, or by displaying formants.

When I do figure it out, I'll update the original post!

I'm glad that many are already finding this useful to analyze pitch! It's been an incredible help to me, and I'm glad to see it helping others as well! Praat is the best! I really wish there was a mobile version so I could test on the go- I'm very frequently wishing I could know my pitch throughout the day by seeing it on a graph instead of a tuner that bounces all over the place.

[Jennygirl]

Okay so in my research I stumbled upon an informative PDF detailing a study on American/French speakers' vocal resonances. Luckily they also differentiate between M/F voices.

Within the article, I found a chart which shows the resonant harmonics or "formants" that make up each vowel sound. What we are looking for is the third formant, or "F3".



In this chart, male F3 falls in between ~2,250hz and ~2,900hz depending on the vowel. Female F3 on the other hand falls anywhere from ~2,600hz to ~3450hz. My speech therapist told me something similar to that with males at 2k-2.5k and females from 3k-3.5k.

F3's are easy to find in Praat. Using the same selection method as with pitch (by dragging over an area), you can select the "Formant" menu and click "Get third formant". You can select all and get an average of the whole thing, or you could check your vowels individually by selecting just a little part and comparing it to the chart above.

In order to be able to "Get third formant", formants need to be shown in the graph. Select the "Formant" menu and click "Show formants"


Then after you've selected an area to analyze:


And it gives you the average F3 result (mine was 2,933.3hz in this example)


I'll be editing the original post with the newness- to keep things concise

Also if anyone has any scientifics to add, please do. I am no expert on this stuff, but I'd love to be taken to school!

[Jennygirl]

I'm finding that I'm getting some inconsistent measurements when analyzing the F3 of a clip in its entirety using these default settings. It only really seems to work where the first 3 formants are clearly formed within the short duration of a single vowel sound. Sometimes they cross over and/or disappear altogether as speech progresses.

This seems to end up making the analysis think that F4 is actually F3.

I was analyzing some deep chested sounding male voices pulled from random youtube videos and getting close to 3,000hz F3's... which should be out of range for male voice. So, something is wrong with that.

When I just selected parts where the first 3 formants were very clear, I kept getting in-range results! So stick to short selections for F3 checking for now until me or someone else figures this out for sure

[Carrie Liz]

Cool beans, Jenny! Thanks for doing all of this research for us!

[anjaq]

Ok. I get about 2800-3000 in that F3 if I use it on small sections (Guess thats androgynous again). It shows these red dotted lines, I try to get parts that have them rather straight, usually those are really short bits only though. I think one has to set the range setting in "pitch settings" to 1500 or so to enable the F3 line to show in the spectrum picture.

[Jennygirl]

For those into the nerdy sorta stuff...

http://clas.mq.edu.au/acoustics/speech_spectra/fft_lpc_settings.html

This page talks some about differences with m/f formants. It's basically the explanation on how a program like Praat works

You can do all sorts of neat things with it. Formants seem to be the key.

F0 - fundamental frequency (or pitch)
F1 - first formant (used to find vowel)
F2 - second formant (used to find vowel)
F3 - third formant (used in finding vocal tract length, or what we call resonance)
F4
F5
F6
and so on, used for finding F0

[Ltl89]

This looks very confusing, so I appreciate you doing everything you can to help the computer illiterate like myself.  Despite the common trans-employment trend, not all of us are computer wizes, lol.  I'll give it a shot once I feel less overwhelmed.   

[vlmitchell]

Thanks for doing this! I'm lazy so, I really didn't want to, per se. :-P

[vlmitchell]

This looks very confusing, so I appreciate you doing everything you can to help the computer illiterate like myself.  Despite the common trans-employment trend, not all of us are computer wizes, lol.  I'll give it a shot once I feel less overwhelmed.

I can make a video tutorial for you later, if you're interested?