Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon) profile picture

he/him. Computational phonetician. Asst. Prof at George Mason University. Opinions my own, not my employer's. @mckphon@lingo.lol

Joined May 2022
Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)Photo by Matthew C. Kelley @mckphon@lingo.lol on Mastodon (@mckphon)
New paper on a gradient approach to speech segmentation! A new feature in the Mason-Alberta Phonetic Segmenter generates gradient intervals for boundary placement in forced alignment. doi.org/10.1515/phon-2025-00…
1
28
Joining the collective sigh of relief among phoneticians for getting my #ICPhS2027 papers submitted tonight 😌😮‍💨
12
I am happy to share that a refreshed version of the #SpeechAccentArchive has been made public today. The search functionality has been reinstated, and many files (but not all) now have plain-text copyable transcriptions.
1
20
Additionally, the source files for the Archive are now available to download through OSF under the CC BY-NC-SA 4.0 license. A page on the site gives more details. (The GIFs are still uploading, so please wait until tomorrow to download if you are interested in those!)
1
18
We've added to the website a free PDF of the Bayesian Workflow book (for non-commercial purposes, like research and exorcism). We put so much work into this book, hope you all find value in it. avehtari.github.io/Bayesian-…
The Bayesian Workflow book is out in print now, just in time to distract you from *gestures all around*. Some links below to code and other resources:
14
161
9
668
74,056
The widths of the boundary regions tend to match phonetic intuitions, where more ambiguous segmentations (e.g., vowel-vowel, approximant-vowel) are wider than less ambiguous segmentations (e.g., fricative-vowel, stop-vowel).
1
17
ALT: Heatmap showing the median widths for different bisegment sequences.
16
New annotation formats are JSONs that list each boundary for a file and its associated widths, and a large CSV with boundaries in each row. Praat TextGrids are also output with PointTiers for the boundaries to aid in visualization.
2
54
ALT: A TextGrid in Praat showing the segmentation overlaid on a spectrogram and a waveform for the first four sounds in "scholar". The boundary region between the low-back vowel and the lateral is wider than the one between the voiceless velar stop and the low-back vowel.
11
And now @ssurovell and @KathyKLTran excluding even more higher ed workers after the second conference committee?? What a farce!
Lines 241-242 of VA's HB 1263 and SB 378 conference report create 2 tiers of grad students. Those paid to teach will not be allowed to bargain, while those paid to research will. This is unjust and needs to be corrected by allowing ALL higher ed workers to bargain!
29
Lines 241-242 of VA's HB 1263 and SB 378 conference report create 2 tiers of grad students. Those paid to teach will not be allowed to bargain, while those paid to research will. This is unjust and needs to be corrected by allowing ALL higher ed workers to bargain!
1
1
1
66
New OA paper, new data set: SingleMALD! We had 40 native speakers of Canadian English do lexical decision for over 26,000 English words and 9,000 pseudowords. Evidence for participants getting faster throughout extensive testing periods. doi.org/10.3758/s13428-025-0…
40
#Gramle 757 3/5 🟩⬛⬛⬛⬛ 🟩🟩⬛🟩🟩 🟩🟩🟩🟩🟩
41
Matthew C. Kelley @[email protected] on Mastodon retweeted
This paper looked at closure duration and f0 perturbation as voicing distinction cues for intervocalic flaps in American English and found that listeners were more likely to make their choice based on the change in f0. doi.org/10.1121/2.0001972 @JanalynAMiklas @GMU_English
3
1
6
194
Paper with @Perry_ScottJ and @B_Tucker_phon on our new #ForcedAlignment tool, MAPS, is out today in Phonetica! doi.org/10.1515/phon-2024-00…
1
4
7
488
Finally, we explore how certain practices used in training these systems is in tension with how phoneticians view the speech signal. Overcoming this tension may require rethinking either what we segment or how we design the targets for our acoustic models.
1
43
A basic end-user system for aligning American English with MAPS is available on GitHub. We will be adding to this system as we continue to develop refinements, new language options, performance improvements, and new tools. github.com/MasonPhonLab/MAPS
39