Christopher Carignan
@carignan.bsky.social
Associate Professor in Speech Science, Director of MSc Language Sciences, DAAPS Lab Director, University College London
One more cohort of PALS0047 R programmers receives their rubber ducks! Everyone say, "Hello World!"
December 9, 2024 at 5:04 PM
One more cohort of PALS0047 R programmers receives their rubber ducks! Everyone say, "Hello World!"
Currently updating the slides for the debugging lecture in my Star-Wars-meme-laden "Programming in R" module to discuss using LLMs in the debugging process. The prompt was to "create a function in R that plots vowels in different colors".
December 4, 2024 at 7:35 PM
Currently updating the slides for the debugging lecture in my Star-Wars-meme-laden "Programming in R" module to discuss using LLMs in the debugging process. The prompt was to "create a function in R that plots vowels in different colors".
Speech science meets Hollywood 🎬
December 3, 2024 at 11:31 AM
Speech science meets Hollywood 🎬
Filming is officially underway for our upcoming promotional video for the SHaPS department and MSc Language Sciences programme!
December 3, 2024 at 10:46 AM
Filming is officially underway for our upcoming promotional video for the SHaPS department and MSc Language Sciences programme!
Sneak peek! Cross-linguistic lip rounding across 86 speakers and three languages shows evidence for a trading relation between protrusion and area, regardless of the phonological status of vowel rounding.
So much time, effort, and computation for the six numbers in this graph!
So much time, effort, and computation for the six numbers in this graph!
August 30, 2024 at 2:33 PM
Sneak peek! Cross-linguistic lip rounding across 86 speakers and three languages shows evidence for a trading relation between protrusion and area, regardless of the phonological status of vowel rounding.
So much time, effort, and computation for the six numbers in this graph!
So much time, effort, and computation for the six numbers in this graph!
This year's PALS0047 intro R programming students have been the absolute best! This nearly made me cry like a baby 😭
December 14, 2023 at 6:26 PM
This year's PALS0047 intro R programming students have been the absolute best! This nearly made me cry like a baby 😭
PALS0047 students are officially code-debugging masters, now that they have the most important, crucial tool in their debugging toolbox... rubber duckies! 🦆🦆🦆
December 13, 2023 at 1:30 PM
PALS0047 students are officially code-debugging masters, now that they have the most important, crucial tool in their debugging toolbox... rubber duckies! 🦆🦆🦆
Advice for the "earbuds method" of measuring nasalance: bandpass filter your signal to approximately the 400-700Hz range before calculating nasalance, regardless of the earbud type used. Incidentally, this is very close to the 300-750Hz filter range already recommended for nasometers!
December 8, 2023 at 4:11 PM
Advice for the "earbuds method" of measuring nasalance: bandpass filter your signal to approximately the 400-700Hz range before calculating nasalance, regardless of the earbud type used. Incidentally, this is very close to the 300-750Hz filter range already recommended for nasometers!
Possibly the strangest image I've yet to include in a paper: me facing off against myself... Spy vs. Spy... Jekyll vs. Hyde... flat vs. silicone earbuds.
November 22, 2023 at 6:04 PM
Possibly the strangest image I've yet to include in a paper: me facing off against myself... Spy vs. Spy... Jekyll vs. Hyde... flat vs. silicone earbuds.
Friends don't let friends consider only global correlations as the basis of measurement validation.
November 22, 2023 at 2:44 PM
Friends don't let friends consider only global correlations as the basis of measurement validation.
Instead of actually working on writing my new paper on validating the "earbuds method" for measuring acoustic nasalance, I'm just making more pretty figures. As is tradition.
November 22, 2023 at 1:06 PM
Instead of actually working on writing my new paper on validating the "earbuds method" for measuring acoustic nasalance, I'm just making more pretty figures. As is tradition.
Look, ma! My nasal cavity made it into a book!
November 21, 2023 at 5:13 PM
Look, ma! My nasal cavity made it into a book!
These have got to be some of the *prettiest* figures I've ever made for a paper 😍
November 16, 2023 at 3:39 PM
These have got to be some of the *prettiest* figures I've ever made for a paper 😍
The Rubber Ducky Army is ready and waiting in my office to help current and future PALS0047 students debug their code
October 26, 2023 at 1:26 PM
The Rubber Ducky Army is ready and waiting in my office to help current and future PALS0047 students debug their code
First day working with a new keyboard and I'm very excited. I'm a simple man, what can I say?
October 16, 2023 at 8:38 AM
First day working with a new keyboard and I'm very excited. I'm a simple man, what can I say?
It's nice to be able to use my principal components variable importance reconstruction (PC-VIR) method, from a previous paper with @egurtzegi.bsky.social, in my current research... even if the methods paper itself was rejected for publication 😬
October 11, 2023 at 5:06 PM
It's nice to be able to use my principal components variable importance reconstruction (PC-VIR) method, from a previous paper with @egurtzegi.bsky.social, in my current research... even if the methods paper itself was rejected for publication 😬
That's one of the reasons I had to redo it! This medical tape I originally used wasn't sticky enough, so I had to get some waterproof athletic tape.
How you're supposed to keep these things staying put in a field setting is beyond me...
How you're supposed to keep these things staying put in a field setting is beyond me...
October 4, 2023 at 3:13 PM
That's one of the reasons I had to redo it! This medical tape I originally used wasn't sticky enough, so I had to get some waterproof athletic tape.
How you're supposed to keep these things staying put in a field setting is beyond me...
How you're supposed to keep these things staying put in a field setting is beyond me...
However, nasalance measures are notoriously unreliable for speech segments that aren't voiced, because the method relies on acoustic radiation. So after further pruning to only *voiced* speech, the performance increases for both flat (R=0.94) and silicon earbuds (R=0.86). 4/8
October 4, 2023 at 2:01 PM
However, nasalance measures are notoriously unreliable for speech segments that aren't voiced, because the method relies on acoustic radiation. So after further pruning to only *voiced* speech, the performance increases for both flat (R=0.94) and silicon earbuds (R=0.86). 4/8
I also needed to ensure that I was only looking at speech data, so pauses were removed. After doing so, flat earbuds correlate surprisingly well with measurements from the nasometer (R=0.86), but silicon earbuds do not (R=0.39). 3/8
October 4, 2023 at 2:01 PM
I also needed to ensure that I was only looking at speech data, so pauses were removed. After doing so, flat earbuds correlate surprisingly well with measurements from the nasometer (R=0.86), but silicon earbuds do not (R=0.39). 3/8
The original proposal for the method includes no acoustic baffle between the channels (as shown here). But in order to compare the measurements directly with a nasometer, a baffle is needed. So the first step was to calibrate the measures against baselines that had no baffle. 2/8
October 4, 2023 at 2:00 PM
The original proposal for the method includes no acoustic baffle between the channels (as shown here). But in order to compare the measurements directly with a nasometer, a baffle is needed. So the first step was to calibrate the measures against baselines that had no baffle. 2/8
After some baseline correction and data pruning, I've now honed on what look to be my final results for this mini-study on validating the earbuds method of measuring nasality against a ground truth of acoustic nasalance. The method *may* work, but with some caveats. 1/8
October 4, 2023 at 2:00 PM
After some baseline correction and data pruning, I've now honed on what look to be my final results for this mini-study on validating the earbuds method of measuring nasality against a ground truth of acoustic nasalance. The method *may* work, but with some caveats. 1/8
Just like last time, the pattern exhibited by the nasometer of convergence towards 50% nasalance as speech amplitude level increases is also exhibited by flat earbuds, but silicon earbuds hover around 50% regardless of the speech level. 3/4
October 3, 2023 at 6:52 PM
Just like last time, the pattern exhibited by the nasometer of convergence towards 50% nasalance as speech amplitude level increases is also exhibited by flat earbuds, but silicon earbuds hover around 50% regardless of the speech level. 3/4
The correlation between flat earbuds and the nasometer is surprisingly good (R=0.79), but the correlation for silicon earbuds is not good at all (R=0.44). Keep in mind that the original article specifically recommends silicon earbuds and discourages flat ones. 2/4
October 3, 2023 at 6:52 PM
The correlation between flat earbuds and the nasometer is surprisingly good (R=0.79), but the correlation for silicon earbuds is not good at all (R=0.44). Keep in mind that the original article specifically recommends silicon earbuds and discourages flat ones. 2/4
The results are in for a more careful, controlled test of the earbuds method for measuring nasality against an established ground truth (a Glottal Enterprises nasometer)...
And the outcome is... the same as before! Don't use them! (at least not as recommended in the original article) 1/4
And the outcome is... the same as before! Don't use them! (at least not as recommended in the original article) 1/4
October 3, 2023 at 6:51 PM
The results are in for a more careful, controlled test of the earbuds method for measuring nasality against an established ground truth (a Glottal Enterprises nasometer)...
And the outcome is... the same as before! Don't use them! (at least not as recommended in the original article) 1/4
And the outcome is... the same as before! Don't use them! (at least not as recommended in the original article) 1/4