iiinova — where your digital self keeps evolving

Miao Zhang @mzhang89.bsky.social · 21/06/2026

I'll be attending LabPhon20 in Montreal from 26.06 to 29.06. Does anyone want to meet up? I'll be presenting an oral presentation on the 27th at 10:45 am.

020

Reposted by Miao Zhang

Chenzi Xu @chenzi.bsky.social · 02/03/2026

The second CorpusPhon Workshop is up, a satellite event at Labphon 2026. Please check this out and submit your abstract! #corpusphonetics #phonetics #labphon2026 #corpusphon

011

Reposted by Miao Zhang

Josef Fruehwald @jofrhwld.bsky.social · 03/02/2026

Chomsky on Foucault: “I'd never met anyone who was so totally amoral” I guess Foucault didn’t retain that distinction forever

2354

Miao Zhang @mzhang89.bsky.social · 15/01/2026

I do have some cool figures that I can share here:

020

Reposted by Miao Zhang

Christian DiCanio @cdicanio.bsky.social · 11/01/2026

Work on Chinese tone/speech errors tends to show that speakers replace entire tones with different ones, e.g. tone /51/ instead of tone /213/. To me that has always meant a kind of holistic melody that is non-decomposable. That’s different from languages where contours are decomposable.

221

Miao Zhang @mzhang89.bsky.social · 15/01/2026

ASA2022 was based on the GAMM chapter and the kinematic analysis chapter, and PCC2023 was based on the duration chapter. I didn't collect new data or perform new analysis.

000

Miao Zhang @mzhang89.bsky.social · 09/12/2025

Tswana is among the languages with a really large VF0 in our data! (language code: tn)!

140

Miao Zhang @mzhang89.bsky.social · 08/12/2025

This Vowel Intrinsic F0 (VF0) pattern shows a deep cognitive bias toward uniform representation of vowels, modulated by flexible, communicative adjustments. Read the preprint here: doi.org/10.31234/osf... #Linguistics #Phonetics #CognitiveScience #VF0 #PhoneticUniversals

doi.org

OSF

020

Miao Zhang @mzhang89.bsky.social · 08/12/2025

🤯 Phonetic Universal Uncovered! 🎤 We analyzed over 60,000 speakers across 75 languages and confirmed a universal phonetic bias: High vowels (like /i, u/) are consistently spoken with a slightly higher pitch (F0) than low vowels (/a/).

doi.org

OSF

2134

Reposted by Miao Zhang

Dr Mircea Zloteanu 🌺🌞🍃 @mzloteanu.bsky.social · 21/11/2025

#statstab #465 How to embrace variation and accept uncertainty in linguistic and psycholinguistic data analysis Thoughts: An accessible paper on communicating your results with nuance. #bayes #bayesian #uncertainty #error #bias #guide #tutorial sites.stat.columbia.edu/gelman/resea...

sites.stat.columbia.edu

032

Reposted by Miao Zhang

Stefano Coretta @scoretta.bsky.social · 15/11/2025

🎉 Finally out in Journal of Phonetics, tutorial with @paulbuerkner.com 📖 "Bayesian beta regressions with brms in R: A tutorial for phoneticians" Accepted manuscript here: doi.org/10.31219/osf... Repo: github.com/stefanocoret... Publisher link: www.sciencedirect.com/science/arti...

media.tenor.com

a close up of a rat looking at the camera with the word drunken written in the corner

ALT: a close up of a rat looking at the camera with the word drunken written in the corner

1298

Reposted by Miao Zhang

Eleanor Chodroff @echodroff.bsky.social · 02/10/2025

Excited to share our new preprint with @mzhang89.bsky.social : “A crosslinguistic corpus phonetic analysis of intrinsic vowel duration” 🎉 🔗 osf.io/preprints/ps...

osf.io

OSF

1112

Reposted by Miao Zhang

Adam L @adam-lg.bsky.social · 20/09/2025

Bring back the iPod classic and the 3.5mm headphone jack

0122

Reposted by Miao Zhang

Adam L @adam-lg.bsky.social · 10/09/2025

Simon Wood, the GOAT of generalized additive models & creator of the mgcv #rstats package, has an Annual Review of Statistics essay on GAMs, available open access #statssky #mlsky www.annualreviews.org/content/jour...

08939

Miao Zhang @mzhang89.bsky.social · 05/09/2025

I feel things corrected by Grammarly feel less AI-generated than those corrected by general AI tools (ChatGPT-like). Is it just my illusion?

000

Reposted by Miao Zhang

Eleanor Chodroff @echodroff.bsky.social · 29/08/2025

🗣️Mozilla Common Voice users!🗣️ Important notice: the client ID does not always correspond to a single speaker ID! Every so often, a single client ID contains more than one speaker’s voice. Our #Interspeech2025 paper examines the extent of this problem and proposes a solution

Interspeech 2025 poster on Quantifying and reducing speaker heterogeneity within the Common Voice Corpus

1163

Reposted by Miao Zhang

Eleanor Chodroff @echodroff.bsky.social · 29/08/2025

✅Similarity scores: huggingface.co/datasets/pac... 📄Paper: www.isca-archive.org/interspeech_... 💻Code: github.com/pacscilab/CV... 💫This was joint work with @mzhang89.bsky.social, Aref Farhadipour, Annie Baker, Jiachen Ma, and Bogdan Pricop

huggingface.co

pacscilab/VoxCommunis at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

021

Reposted by Miao Zhang

Darin Flynn @phono-logical.bsky.social · 26/08/2025

LabPhon 20 will be held in Montréal June 25–28, 2026, on the theme “Looking Back and Looking Forward,” to reflect on the field’s foundational contributions while highlighting new directions in laboratory phonology. Abstract submission deadline: Dec 1, 2025 labphon.org/labphon20/home

labphon.org

Home | Labphon

044

Miao Zhang @mzhang89.bsky.social · 21/08/2025

The similarity score file can be found in our VoxCommunis huggingface repo: huggingface.co/datasets/pac.... You can also see the scripts we used to obtain the similarity scores here: github.com/areffarhadi/...

huggingface.co

pacscilab/VoxCommunis · Datasets at Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

020

Reposted by Miao Zhang

Josef Fruehwald @jofrhwld.bsky.social · 16/06/2025

Introducing the tidynorm package! It's got convenience functions for applying your favorite vowel normalization methods to point measures, formant tracks, and DCT coefficients in a tidyverse workflow, as well as a flexible framework for defining your own normalization methods!

jofrhwld.github.io

Introducing tidynorm – Væl Space

Here’s a brief introduction to the new tidynorm package.

24415

Miao Zhang @mzhang89.bsky.social · 10/06/2025

When people talk about neutralization in phonology, it's very important to check some phonetic data. It's very probable that we either didn't perceive it or overinterpreted some variance as non-natives.

020

Reposted by Miao Zhang

Posit @posit.co · 09/06/2025

ggplot2 is turning 18! 🎂 For nearly two decades, it’s helped data scientists turn complex data into clear, beautiful insights. We’re throwing a birthday party at Data+AI Summit, with treats and limited-edition swag. Come celebrate with us and @hadley.nz! 📍 Posit Lounge (402) 📅 June 10, 6–8pm

611916

Miao Zhang @mzhang89.bsky.social · 31/05/2025

The worst kind of echo chamber.

020

Miao Zhang @mzhang89.bsky.social · 27/05/2025

The debate doesn't exist in China, we just call it [ʈʂi˥ aɪ˥ ɛ˧˥fu]

000

Miao Zhang @mzhang89.bsky.social · 27/05/2025

In case people don't use it very often, or never knew its existence, glimpse() from dplyr is a much better function to use when you want to have a very rough look at your dataset than head() or summary().

120

Miao Zhang @mzhang89.bsky.social · 26/05/2025

Congratulations! Wow starting from Associate Professor sounds really great!

110

Miao Zhang @mzhang89.bsky.social · 26/05/2025

The only upside of OSF compared to Huggingface is that you get to have a doi link on OSF, which makes your repo look slightly more professional. But all the other aspects are way worse. Huggingface provides so many useful tools to manage your data/model, making OSF almost a joke in this regard.

010

Miao Zhang @mzhang89.bsky.social · 26/05/2025

And storage limit. We migrated our data repo to Huggingface.

120

Reposted by Miao Zhang

Rasmus Puggaard-Rode @rpuggaardrode.bsky.social · 14/01/2025

💣 praatpicture version 1.4.0 on CRAN! 💣 (1/3)

25215

Miao Zhang @mzhang89.bsky.social · 22/05/2025

I feel Google Scholar PDF is better cuz you can get way more stuff in the pop-up window in Google Scholar PDF than the one in Zotero reader, including the link/citation/pdf of the references.

120

Reposted by Miao Zhang

Chris Offner @chrisoffner3d.bsky.social · 21/05/2025

I also use Zotero for most of my reading. Other than that, the Google Scholar PDF Reader extension is the best thing w.r.t how citations are handled: chromewebstore.google.com/detail/googl...

chromewebstore.google.com

Google Scholar PDF Reader - Chrome Web Store

Supercharge your paper reading: follow references, skim outline, jump to figures, cite and save.

211

Miao Zhang @mzhang89.bsky.social · 21/05/2025

Wow, I just checked if there is a newer version, literally just 1 hour ago.

110

Reposted by Miao Zhang

Rasmus Puggaard-Rode @rpuggaardrode.bsky.social · 21/05/2025

Very exciting news for PraatSauce users! We've just pushed a new version which fully rewrites the code base, making things faster and simpler to use. Instead of settings parameters in a bunch of Praat windows, you now set them in a spreadsheet file that looks like this github.com/kirbyj/praat...

1112

Miao Zhang @mzhang89.bsky.social · 21/05/2025

All the score and the clean file list will be shared online by our team later. 4/4

010

Miao Zhang @mzhang89.bsky.social · 21/05/2025

As a result, we were able to obtain a score of 0.38 as the threshold of rejecting the same speaker hypothesis. This score is close to that of the ResNet-293 model trained on English data. Overall, less than 10% of the recordings in our dataset from 75 languages were under 0.38. 3/n

110

Miao Zhang @mzhang89.bsky.social · 21/05/2025

In our paper, we used a speaker verification system, ResNet-293 model trained on VoxBlink2 dataset to obtain a similarity score of the recordings within each client ID. We then employed an auditing process by five of the authors to evaluate speaker heterogeneity in different similarity scores. 2/n

110

Miao Zhang @mzhang89.bsky.social · 21/05/2025

Mozilla Common Voice datasets are very useful multilingual speech corpora for both speech technology and phonetic analysis. While it provides a client ID as an approximation of true speaker id, but due to its crowdsourcing nature, multiple speakers can contribute under the same client ID. 1/n

110

Miao Zhang @mzhang89.bsky.social · 21/05/2025

I’m thrilled to announce that our paper, Quantifying and Reducing Speaker Heterogeneity within the Common Voice Corpus for Phonetic Analysis, coauthored with @echodroff.bsky.social, Aref Farhadipour, Jiachen Ma, Annie Baker and Bogdan Pricop from @cl-uzh.bsky.social was accepted for INTERSPEECH2025.

2101

Miao Zhang @mzhang89.bsky.social · 13/04/2025

I personally feel it should depend on the tonal phonology of the language

000

Miao Zhang @mzhang89.bsky.social · 13/04/2025

I happen to always use car::Anova(). Thanks for clarifying this!

010

Reposted by Miao Zhang

Matthieu Boisgontier @matthieuboisgontier.com · 13/04/2025

When running ANOVAs in #R, use car::Anova(). aov() and anova() use Type I sums of squares, meaning that order matters, which can distort results in unbalanced designs. car::Anova() is safer because it uses Type II sums of squares by default), each effect is adjusted for all the other effects.

3114

Miao Zhang @mzhang89.bsky.social · 13/04/2025

When you train an acoustic model for a tone language with tone labels in the phone set, do you keep the underlying tone label or the surface tone label?

100

Miao Zhang @mzhang89.bsky.social · 28/03/2025

I feel that the world I was familiar with before the pandemic will never return.

020