Self phonetic survey

[Qwynegold]

I decided to examine my own speech by doing recordings and looking at it with Praat. Right now I'm examining my Finnish speech, but I might do Swedish later. There's two reasons for doing this. Firstly there are some things I would like to find out about Finnish allophony. Secondly, when I'm working with a conlang with phonemes that I can't confidently pronounce, it helps if I have some "normal" phonemes to compare my conlang phonemes with.

If there are other Finnish speakers here that would like to participate, you can make your own recordings and then maybe we can combine our data, so we can see if there are some things that are peculiar only to my speech, or if they're common across more speakers.

Or if there is anyone that has examined their own speech with a speech analysis program in any language, and want to share their results, you can do that here.

[Qwynegold]

First I examined vowel length. I wanted to see what effect stress has on length. I made the following word list:

Code: Select all

kapipa		taloon		pataapa
kaataapa	aateko		Aatuko
petipä		ketoon		retiisi
teekö		teettääkö	Eevako
pitääpä		likööri		lipsuupa
kokohan		kotiisi		motiivi
kootako		kookos		koostumus
pupupa		kutooko		tupako
kuuko		uutuudet	puutuuko
kypärä		pystyypä	ryttyyn
kyykö		hyytyypä	pyyntöön
tätikö		kädellä		tätäpä
tökötti		kökköön		ötökkä
kiitos		piisaako	kiitääpä
päätekö		päättääkö	näätäkö
töötätä		söötti		lööpit
taiteet		pateet		aikeet
tykyttää	tytöt		kehykset	päädyt

Now this might not have been the best way to go about doing things. I just read a list of words, instead of reading sentences, and some of the vowels are separated by other consonants than plosives.

For stressed vowels I took the vowel from the first syllable, and for unstressed vowels I took the one from the second syllable. I'm not sure how exactly the rules for secondary stress work like in Finnish, but I think this should guarantee that all vowels have the stress I think they have. So I have three examples for each kind of stressed-short, stressed-long, unstressed-short and unstressed-long vowel. I think that ideally one is supposed to have at least five examples for each category, but that would've been too much work. :/



In this diagram, the blocks are divided into three parts. The lowest one shows the shortest length found among the examples, and the highest one shows the longest length found among the examples. The middle part (the black line between the two colors) shows the average length. The dark line labled at the far right side of the diagram show the average length of stressed short vowels, unstressed short vowels, unstressed long vowels, resp. stressed long vowels when you combine all vowel qualities.

What I found out from this is that the difference between short and long is more pronounced in stressed syllables than in unstressed syllables. I.e. short vowels become even shorter when they're stressed, and long vowels become slightly longer when stressed.

I also found out that when you look at stressed short vowels vs. stressed long vowels, or unstressed short vowels vs. unstressed long vowels, values never over-lap (the colored portions of the blocks are at completely different heighths), except for two cases: unstressed [o] vs. [oː] and unstressed [æ] vs [æː]. I can't tell why this happened. But length is mostly unambiguous.

Oh, and when stressed long vowels are 2.98 times longer than short vowels on average; and if unstressed, long vowels are 1.841 times longer than short vowels.

[missals]

I love this! Great data visualization. I did a study on "Canadian raising" e.g. rider [ɹaɪɾɚ] vs. writer [ɹʌɪɾɚ] in some American English speakers and found some similar patterns, i.e. that the difference between the two was exaggerated in primary-stressed syllables and "dampened" in secondary-stressed syllables in a rather complex manner.

It's always so interesting to see how phonological distinctions like this are actually implemented phonetically. I have no idea if there are any gaps in the literature on Finnish vowel length that need to be filled, but if you did this on some other people I bet this could be published or presented at a conference or something like that.

[Qwynegold]

I love this! Great data visualization.

[...]

It's always so interesting to see how phonological distinctions like this are actually implemented phonetically. I have no idea if there are any gaps in the literature on Finnish vowel length that need to be filled, but if you did this on some other people I bet this could be published or presented at a conference or something like that.

OMG thanks!

I did a study on "Canadian raising" e.g. rider [ɹaɪɾɚ] vs. writer [ɹʌɪɾɚ] in some American English speakers and found some similar patterns, i.e. that the difference between the two was exaggerated in primary-stressed syllables and "dampened" in secondary-stressed syllables in a rather complex manner.
Those two words are different??!

[Qwynegold]

This is each vowel quality regardless of length and stress. We can see that some vowels have quite a lot of variation, while others have much less. The weird thing is that /ø/ overlaps with /y/, and /o/ with /u/.

I overlaid this with the vowel triangle that's supposed to replace the old vowel rectangly thing. Without distorting the proportions, this vowel triangle doesn't fit that well. If I make this triangle fit the height of my vowels, the back and front vowels go way outside the triangle. If I try to match the width instead, then I have no vowels in the low or high regions.

The two (kind of) vertical lines in the middle are supposed to stand for front rounded and back unrounded vowels. My front rounded vowels do match this set up though, as they appear to be much more back then their unrounded counterparts.


These are the vowels separated by length and stress. There are a few really weird datapoints. [ˈɑ] and [ˈɑː] both have one instance that is extremely high, [æ] has one instance that's almost in the region of [ˈe], and [ø] has one instance that's higher than any /y/. I don't know what to make of this. Maybe I had gotten better data if I had recorded more instances of each vowel.

[ˈo] overlaps a lot with [ u]. This is odd. Otherwise I can't really draw any conclusions from this about whether length or stress has any effect on vowel quality. Does anyone else see any patterns?

Here are some other questions and things I noticed while working with this. I will probably not investigate these things further, but maybe they can serve as inspiration for someone?

• If a syllable is repeated, e.g. koko, pupu, then does the first vowel become extremely short? It looked to me like there could be such a trend.
• It seemed as if a long vowel in a monosyllabic root will have more extreme F1/F2 (even if that root is inflected to become a polysyllabic word), than a long vowel in a polysyllabic root. Is this observation true?
• Vowel-initial words seemed to sometimes get a ʔ at the beginning. But I didn't observe any word-final ʔ. Though then again, I'm not sure how one would even detect a word-final ʔ.
• Interestingly, I found one instance of an unvoiced vowel. (I didn't record any data from this word.)

[Qwynegold]

Next, I'm going to investigate allophones of /h/. But I'm wondering, how do I tell different non-sibilants apart in a spectrogram/waveform? Also, is it possible to see if an [h] is "pure phonation h" or an "h with friction"? From what I've seen so far, [h] is often completely invisible.

[Creyeditor]

This is each vowel quality regardless of length and stress. We can see that some vowels have quite a lot of variation, while others have much less. The weird thing is that /ø/ overlaps with /y/, and /o/ with /u/.

I think this is common in languages where the mid vowels are high-mid. IIRC, F3 is a helpful cue to distinguish between these in some languages.

The two (kind of) vertical lines in the middle are supposed to stand for front rounded and back unrounded vowels. My front rounded vowels do match this set up though, as they appear to be much more back then their unrounded counterparts.

Keep in mind that F2 and F3 correlate. The more rounded your rounded front vowels are, the more "back" they will appear in the chart.

These are the vowels separated by length and stress. There are a few really weird datapoints.

I think this is a typical result for small-scale survey with a few datapoints, at least in my experience. Don't worry too much about it. Maybe you could do two additional charts where you only separate stressed/unstressed and long/short, respectively?

Next, I'm going to investigate allophones of /h/. But I'm wondering, how do I tell different non-sibilants apart in a spectrogram/waveform? Also, is it possible to see if an [h] is "pure phonation h" or an "h with friction"? From what I've seen so far, [h] is often completely invisible.

I don't know of an easy way to detect /h/. Maybe formants become less visible, but there is no pause (as there is in glottal stops). Maybe you have to look at shimmer and jitter to detect breathy voiced parts? I never tried any of these though.

[Qwynegold]

This is each vowel quality regardless of length and stress. We can see that some vowels have quite a lot of variation, while others have much less. The weird thing is that /ø/ overlaps with /y/, and /o/ with /u/.

I think this is common in languages where the mid vowels are high-mid. IIRC, F3 is a helpful cue to distinguish between these in some languages.

The two (kind of) vertical lines in the middle are supposed to stand for front rounded and back unrounded vowels. My front rounded vowels do match this set up though, as they appear to be much more back then their unrounded counterparts.

Keep in mind that F2 and F3 correlate. The more rounded your rounded front vowels are, the more "back" they will appear in the chart.

I didn't measure F3, because I thought "nah, it will be too much work. F1 and 2 will probably be enough."

These are the vowels separated by length and stress. There are a few really weird datapoints.

I think this is a typical result for small-scale survey with a few datapoints, at least in my experience. Don't worry too much about it. Maybe you could do two additional charts where you only separate stressed/unstressed and long/short, respectively?

Ah yeah, I should probably do that.

Next, I'm going to investigate allophones of /h/. But I'm wondering, how do I tell different non-sibilants apart in a spectrogram/waveform? Also, is it possible to see if an [h] is "pure phonation h" or an "h with friction"? From what I've seen so far, [h] is often completely invisible.

I don't know of an easy way to detect /h/. Maybe formants become less visible, but there is no pause (as there is in glottal stops). Maybe you have to look at shimmer and jitter to detect breathy voiced parts? I never tried any of these though.

Okay. I'll have to see what it looks like once I have recorded.

[Creyeditor]

I didn't measure F3, because I thought "nah, it will be too much work. F1 and 2 will probably be enough."

I think this is totally okay. F3 is often harder to measure and interpret anyway. It also disorts the nice two-dimensional chart

[Kuchigakatai]

The beautiful graph you made doesn't load anymore, without failing immediately either, for some reason:
https://postimg.cc/sBz1D4g9 <- your link

Could you re-post it somewhere else?

[Qwynegold]

I think this is totally okay. F3 is often harder to measure and interpret anyway. It also disorts the nice two-dimensional chart

Ikr

The beautiful graph you made doesn't load anymore, without failing immediately either, for some reason:
https://postimg.cc/sBz1D4g9 <- your link

Could you re-post it somewhere else?

Only the first graph? Hmm, I can see it fine and that link also works for me. I have it on my Dropbox: https://www.dropbox.com/s/n1e3cyfialqzn ... t.png?dl=0

[Kuchigakatai]

Only the first graph? Hmm, I can see it fine and that link also works for me.

Thanks, and yes, I was interested in that first graph. It didn't work at all at the time of my post, but now it's working again.

[Qwynegold]

Only the first graph? Hmm, I can see it fine and that link also works for me.

Thanks, and yes, I was interested in that first graph. It didn't work at all at the time of my post, but now it's working again.

Ah, okay. Maybe the site was down at that moment or something.

Btw, I realized that I have already done some examinations of my Swedish speech in a class once. I should see if I can find my old stuff, and if there was anything interesting there, and maybe post it here.