To properly view the phonetic symbols in the text below, you must have installed either SILDoulos IPA93 or Lucida Sans Unicode. If you are desperate to see phonetic symbols in SIL Sophia or SIL Manuscript, or some other kind of Unicode, drop me a line.

Solution for December 2002


"Silk flowers have no smell."

Segmental cues

Lower-case S, IPA 132
SIL [s], Unicode [s]
We've got a fair bit of noise. It's not remarkably loud, so I guess I won't go on too long about how much louder sibilant noise is than channel noise. But starting about 100 msec and continuing to just past 250 msec, there's frication. Notice the 'profile'. There's some noise at every visible frequency (more as the fricative goes on), but the noise is definitely concentrated off the top of the spectrogram. This is absolutely characteristics of /s/, i.e. alveolar (coronal) fricative, and it's due in part to the jet of air being blown against the upper teeth. Sibilant fricatives are all about jet management. Remember that, next time you're haranguing a six-year-old. ;-)

Small Capital I, IPA 132
SIL [I], Unicode [ɪ]
This vowel is tough, and I probably shouldn't have transcribed it so directly. There's a heckuva lot of movement here, due in part to the following consonant. The F1 is just a little low of neutral (neutral for F1 is somewhere around 500 Hz for my voice). So this vowel is a little high of mid. I decided this because there seem to be something like four strong harmonics, at roughy 200, 400, 600 and 8oo Hz. Notice that the bottom two are stronger and smear together more. So I'm assuming the peak of the filter function (the first resonance) is somewhere between the two. The peak of the filter function gives the most boost to the underlying harmonics, and does the most smearing. Which is why I think the peak is low of mid rather than somewhere else. The F2 starts just a tad high of the neutral F2 range (1500-1600 Hz) or so, but falls rapidly. Since this doesn't look like a backing diphthong (i.e. [aw]), I'd say this was transitional. Transitional to what we'll discuss later. At that left edge of the vowel, the F1 indicates a higher-than-mid vowel, and a slightly front vowel, with a basically neutral F3. So what vowel is slightly high and slightly front. Well, something like small-cap I.

Lower-case L + Mid Tilde, IPA 155 + 428
SIL [lò], Unicode [ɫ]
Segmenting off this bit depends on noticng that the harmonics above 1500 Hz or so die off, but the voicing and lower harmonics (supported by F1 and F2) stick around. So something is happening here which isn't happening in the 'vowel' segment earlier. That might be just voice-quality or pitch change zapping the upper frequencies, but it's better to assume that there may be something here and be wrong than to assume there isn't. Just cuz disproving a positive is easier than proving a negative. Okay, so if there's something going on here, what is it? The F1 is low. The F2 is low, but that F3 is outrageously high. /l/ in my dialect is pretty much always dark (low F2), and is almost always accompanied by a high F3 (or occasionally an F3 that looks neutral and a high F4). Low F2, high F3, must be dark /l/.

Lower-case K, IPA 109
SIL [k], Unicode [k]
There's almost no evidence of a velar here. I'm looking at a gap from about 400 msec or so to about 480 msec or so. But look at that. There's a burst amid some light noise, and then right after 500 msec there's another clunk in the same range. Could just be a clunk, but that timing, 30 msec or so between clunks, just looks like a double-burst, characteristic of velars (and occasionally laminal coronals). Which is pretty much the only evidence that this could be velar. There's no transitional information available in the preceding voiced bit, and there fricative noise following is really unhelpful. But that double burst. And even that's not really great looking, since it seems to be in the F1 range, and we really like the burst of a velar to be in the F2 and F3 range. But, here's where you have to consider context. So we'll go on to the following context, and then come back to this.

Lower-case F, IPA 128
SIL [f], Unicode [f]
Noise, so a fricative. Pretty clearly voiceless, so voiceless. Very strong. And apparently concentrated in the uppre frequencies, but much less obviously than the initial /s/ in this utterance. Note the absolute lack of any frequency-continuity with th formants. Compare that with the /s/, where there's not a lot, but at least you could argue that the noise is sort of close to where the F2 and F3 transitions are. But there's just no continuity here. So I'd say this fricative is unfiltered, i.e. the noise has to be coming from the front (lip-end) of a mostly closed vocal tract. Otherwise, some cavity somewhere should be offering a resonance. And it's not. There being no bilabial fricatives (normally) in my speech, the next best choice is something dental. Either labio- or inter-.

Okay, so if we say interdental, we might say the double burst is due to laminality. But I have mis-spoken. It's not laminality per se, but the length of the closure along the hard palate. And the upper teeth don't provide that kind of length. So let's say labiodental. Then we could have a velar stop which would explain the double burst, and we could have the low-frequency burst, since it has to resonate through the approach to the labial 'closure'. Voilá, everybody's happy.

Lower-case L + Mid Tilde, IPA 155 + 428
SIL [lò], Unicode [ɫ]
Remember what I said before about low F2, and high F3 OR low F2, neutral F3 and a high F4? Well, here's the high F4 part of that. F3 is just dead flat through this vowel, but the F4 starts up somewhere around 3999 Hz! That's just freaky. It's also uncontestably high. Dark /l/.

Lower-case A, IPA 304
SIL [a], Unicode [a]
Had some trouble deciding how to transcribe this one. I've got intuitions about what the structure of this word is, and none of them are borne out by this spectrogram. So, in the end, I did the following. I abstracted away from the /l/ transitions, so I started looking at about 625 msec or so. Then I notice there's sort of a 'hump. in F1 and F2 between 625 and about 725. After 725 there's something else going on. There's very little evidence of diphthongization in this stretch, depending on what you think is going on after 725 msec. Okay, so we've got something that's quite low. It's strikingly back, though not as back (or round) as either the /l/ or the following bit. So I chose /a/. Which should technically be used for a front vowel. But this vowel just doesn't feel back. So I cheated.

Lower-case W, IPA 170
SIL [w], Unicode [w]
The F1 is ambiguous here, but the F2 definitely has a dip about 750 or so. So something not at all low (but not particularly high) and backish/roundish. Fully voiced and sonorant, by the way. Non-nasal (no zeroes). And back. Could be another /l/, although it would be nice if the F3 weren't taking a nosedive and F4 were visible. Then I notice the following transition (from about 775 msec to the F2 offset at 850msec or so). It's straight. Transitions are usually curved, i.e. the velocity of the movement slows as it approaches the target. The only thing that I know of that has consistently straight, slow F2 transitions is /w/ across a following vowel. Good guess. Who knows if it's right?

Turned R + Syllabicity Mark, IPA 151 + 431
SIL [¨`], Unicode [ɹ̩]
Low F3, reaches a minimum just about 800-825 msec, around 1800 Hz or so. Nuff sed?

Lower-case Z + Under Ring, IPA 133 + 402
SIL [z8], Unicode [z̥]
This doesn't look voiced. Except for the teeny hint of energy around 300 Hz, which just might be a hint of a harmonic. But there's no striations in the voicing bar, I don't think. So who knows. This looks like a fricative, and judging from the high frequency energy, it looks alveolar and sibilant. But weak. I've transcribed it as a devoiced [z], which is not quite the same thing as a voiceless [s], since voiceless implies vocal fold abduction, where devoiced just implies lack of vibration. So the relative weakness of the energy here (which implies low airflow) suggests something that is 'underlyingly' voiced. That's my story and I'm sticking to it.

Lower-case H, IPA 146
SIL [h], Unicode [h]
This is brilliant. It's incredibly loud, so we might think sibilant (I'm assuming I don't have to point out that this is obviously frication). But look at those formant-like bands. I'm still trying to work out whether something aperiodic can have formants, or should have resonances, or whatever. But look. Very obviously F2 moving through this stretch of frication, and F3 pretty flat. Some organization in the F4 range, and if you work at it, you can see some organization in the F1 range, which is high (there's a harmonic in the following vowel at about 600 and another between 900 and 1000 Hz (which is consistent with the pitch track, which indicates the pitch here is just a hair above 150 Hz). Anyway, those two strong harmonics look like F1 to me, so if you follow those frequencies back into the fricative we're discussing, voilá organization.

Ash, IPA 325
SIL [Q], Unicode [æ]
High F1 (low vowel), neutral F2 (not amazingly front, but not at all back). At least this one has a higher F2 than the previous one. So whatever you called that one, call this one slightly fronter.

Lower-case V, IPA 129
SIL [v], Unicode [v]
Well, it's fricative. Weak but fricative. There's more of that energy where the voicing bar should be, and notice the interruption of the background noise, suggesting really, really weak voicing, even if it doesn't look amazingly striated. So weakly voiced, weak fricative. So how do we identify the fricative. It's not sibilant. It doesn't have the filtered-organizational quality of /h/. Which leaves the front fricatives (labiodental and (inter)dental). Not a lot of transitional information, but if you use your imagination--er, I mean insight, the F3 transition into this fricative may be dropping a little. A little. And if you think that's true, then you can convince yourself that F2 is doing it as well. And then you have labial-looking transitions. So there you go.

Lower-case N, IPA 132
SIL [n], Unicode [n]
Ah, something obviously voiced, but weak compared to the following vowel, and with a nice obvious zero in it. The sharp transition between this and the following vowel, along with the zero (about 1000 Hz, where all the energy gets zapped out of the spectrum), suggest nasal more than anything else. Oral approximants can be similarly weak, but the transitions (both in terms of resonance and just the amplitudes) are usually just a little bit smoother. So Nasal. If you know my voice, you know that 1500 Hz for a resonance is usually a pretty good indicated of alveolar closure. (My bilabial nasal has a resonance closer to 1000 Hz.) The other voice is velar, but the transitions in the following vowel don't look velar at all.

Lower-case O + Upsilon, IPA 307 + 321
SIL [oU], Unicode [oʊ]
Middish vowel, perhaps rising at the end (F1 starts at about 600 Hz, and possibly falls a little starting around 1400 msec). Backish vowel (F2 starting just below 1500 Hz) and getting backer (or, more likely, rounder) over the course of the vowel. So middish and sort of back, getting higher and backer.

Lower-case S, IPA 132
SIL [s], Unicode [s]
'nother one.

Lower-case M, IPA 114
SIL [m], Unicode [m]
There's 75 msec or so of gap in the upper frequencies, but if you follow it down to the low frequencies, you see it's clearly voiced (striations from about 1600 on). So low amplitude and voiced. Must be a sonorant. Following /s/ (unless you believe the preceding word is [noUs]) so it must be a nasal (except eng) or an approximant (except /r/). Doesn't look like an /l/ or a /w/ (not enough lowness to the onset of the F2), and the lack of energy even below 1000 Hz is suspiciously nasal-zero looking. Transitions could be bilabial (F2 wouldn't start that low if it were alveolar or velar--at least not usually) so /m/ is the best guess.

Epsilon, IPA 303
SIL [E], Unicode [ɛ]
Okay, here's where it all goes out the window. This is a lowish vowel. Not as low as low gets, but the F1 is definitely high of the mid-range. It gets lower, which is a little weird, but oh well. The F2 looks a whole lot like the preceding vowel, but I promise there's nothing round (at least) about this vowel. Which just goes to show you how 'tonality' works. The F3 looks a little low, but I'm willing to bet that's just bilabial-transition low rather than rhoticity low. So we're looking for a lowish vowel. We'll come back to this one. after we do the final sound.

Lower-case L + Mid Tilde, IPA 155 + 428
SIL [lò], Unicode [ɫ]
Okay, there's this moment, around 1850 msec where the harmonics and formants lose their energy. It's not quite where I placed the segment break, since I was looking more at what I thought was going on in F2, where it looks like where it starts to fall apart. But whatever. After that moment, we've got a very high F1 (don't ask me why), and a very, very low F2. So this is as back as it gets. The F3 doesn't tell us a lot, except it's just a little bit high.

Okay, so if the vowel is /o/ and this is an off-glide (like a /w/), this sentence doesn't make any sense. If this is a dark /l/, then the word could either be 'small' 'smole' or 'smell'. Which is the most likely to go with the 'silk flowers'?

So (working backwards from the hypothesis that this word is 'smell'), what this is is a very dark /l/, so dark it's velarizing/backing (slightly) the (underlyingly not-particularly front, low-of-mid vowel /E/ which precedes it. (There are people who definitely have a highish /e/ vowel in this word. I'm not one of those people.) Monosyllabic words with final /l/ are notorious for goofing around with the vowel quality in North American English.

The prosody and intonation

Once again, I've played with the current E_ToBI transcription conventions. Rather than a separate orthographic tier, I've aligned the Break Indices to the segmental transcription. I've combined the Tonal Tier and the Break Index Tier as a single line. I align word-level (*) tones with the middle of the marked vowel, but phrase accent (-) tones and boundary (%) tones to the left of their appropriate Break Index.

"silk"
Break Index: 1
This word is receiving some kind of focus. silk flowers versus real flowers, I guess. As I understand the current conventions, lexical words receive (at least) a 1BI and a *-tone, unless deaccented in some way. Receiving contrastive focus, this word receives an H*. If this were a normal declarative. It might receive an L*.

"flowers"
Break Index: 2
Okay, the way I read the conventions, 2BIs are for a) something which 'feels' like a phrasal break, but doesn't receive any particular tonal or prosodic mark (i.e. no phrase-final lengthening). This is different from a ?BI, which indicates an apparent mark, but of uncertain or ambiguous BI level. So I gave this a 2, just because it seems a little bigger than a 1, but there doesn't seem to be any mark. Lexical word, so I gave it an L* since it is near the apparent baseline. In a normal sentence, this would be H*, but due to the focus of the preceding word, this is L*.

"have"
Break Index: 1
Another word, so it gets a tone. I've used H* because it feels like a high, although it is obviously lower than the first one. I don't think there's a strong or obvious contrast between this intonational contour and one in which the HiF0 is here rather than on the previous H*. Give it a try. If this were a normal rather than focus sentence, this would be a L*.

"no"
Break Index: 1
Since 'no' isn't lexical in the usual sense, I'd be tempted to give this a 0BI, but since it seems to have a distinct L*, I gave it a 1. There's a certain amount of interpolation between the preceding H* and the pitch minimum on this vowel. On the other hand, I think the pitch track indicates a drop to baseline here, rather than something neutral between the preceding H* and the Ls to come. In a normal sentence, this could be either H* or L*, or unaccented.

"smell"
Break Index: 4
Low throughout, so I gave it a L*. The slight rise in pitch at the left edge of the [m] voicing I think is just a microperturbation due in part to the voicelessness of the preceding fricative. (Voiceless sounds often are accompanied by a slight increase in pitch. Supposedly this has to do with tightening of the folds that accompanies or anticipates abduction.) The utterance-level 4BI requires (as I understand the current conventions) a phrase accent and a boundary tone. Since there's no evidence of anything but L throughout, that's what I've selected. If there were a change from H to L here, or the reverse, I'd have a big trouble deciding between the H* L-L% and H* H-L% or whatever. But no such conflict here.