iiinova — where your digital self keeps evolving

Kwanghee Choi @juice500ml.bsky.social · 10/09/2026

3 papers submitted & accepted at #SLT2026 🎉 Great to see that phonological vector arithmetic is indeed effective across several new applications. Bunch of computational linguistics this time, see you in Palermo!

031 0

Kwanghee Choi @juice500ml.bsky.social · 07/04/2026

4 papers submitted & accepted at #ACL2026 🎉 So grateful to work alongside & learn from amazing minds, pushing the boundaries of speech technologies, machine learning, and computational linguistics. See you in San Diego!

042 0

Kwanghee Choi @juice500ml.bsky.social · 16/03/2026

📄 Paper 2 (submitted to IS): "Self-Supervised Speech Models Encode Phonetic Context via Position-dependent Orthogonal Subspaces" We further show how sequences of phone(me)s can be encoded, i.e., contextualize, in a single S3M frame. arxiv.org/abs/2603.12642

110 0

Kwanghee Choi @juice500ml.bsky.social · 16/03/2026

📄 Paper 1 (submitted to Jan ARR): "[b] = [d] − [t] + [p]: Self-supervised Speech Models Discover Phonological Vector Arithmetic" We show how phone(me)s are encoded in S3Ms: as a linear combination of phonological feature vectors. arxiv.org/abs/2602.18899

100 0

Kwanghee Choi @juice500ml.bsky.social · 16/03/2026

𝐒𝐞𝐥𝐟-𝐬𝐮𝐩𝐞𝐫𝐯𝐢𝐬𝐞𝐝 𝐒𝐩𝐞𝐞𝐜𝐡 𝐌𝐨𝐝𝐞𝐥𝐬 𝐚𝐫𝐞 𝐏𝐡𝐨𝐧𝐨𝐥𝐨𝐠𝐢𝐜𝐚𝐥 𝐕𝐞𝐜𝐭𝐨𝐫 𝐌𝐚𝐜𝐡𝐢𝐧𝐞𝐬! 🗣️ Excited to be giving an invited talk this Thursday (March 19th, 3pm Amsterdam time)! Huge thanks to @mdhk.net at University of Amsterdam for the invite 🙏

262 0

Kwanghee Choi @juice500ml.bsky.social · 15/08/2025

There's also bunch of engineering tricks that can improve the performance. We provide a pareto-optimal baseline after applying all the available tricks, positioning our work as a foundation for future works in this direction. github.com/Masao-Someki... (9/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 15/08/2025

We also verified that DSUs are learnable with smaller weights (# of layers), i.e., more lightweight! This implies that we're using self-supervised models inefficiently when extracting DSUs. (8/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 15/08/2025

We verified that DSUs are learnable with limited attention size (window size), i.e., streamable! This implies that DSUs are temporally "local". (7/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 15/08/2025

After modifying the architecture, we fine-tune it with the DSUs extracted from the original full model. We're now understanding DSUs as "ground truth" for smaller models. (6/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 15/08/2025

However, the underlying Transformer model is heavy and non-streamable. We make the model more lightweight (via reducing # of layers) and streamable (via streaming window). (5/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 15/08/2025

A whirlwind overview of discrete speech units (DSUs): we first train a Transformer model with self-supervision (i.e., self-supervised speech models, S3Ms). Then, we simply apply k-means on top of it. Then, the k-means cluster indices becomes DSUs! (3/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 15/08/2025

Can we make discrete speech units lightweight🪶 and streamable🏎? Excited to share our new #Interspeech2025 paper: On-device Streaming Discrete Speech Units arxiv.org/abs/2506.01845 (1/n)

211 0

Kwanghee Choi @juice500ml.bsky.social · 29/04/2025

Check out my presentation and poster for more details. I'll see you at NAACL, 4/30 14:00-15:30 Poster Session C! youtu.be/ZRF4u1eThJM (9/9)

010 0

Kwanghee Choi @juice500ml.bsky.social · 29/04/2025

We provide an extensive benchmark containing both pathological and non-native speech, with 8 different methods and 4 different speech features. It measures how well does the speech features model each phonemes accurately. (7/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 29/04/2025

So, why is allophony important? We have to model each phonemes accurately for the atypical speech assessment task. It has direct applications to non-native and pathological speech assessment. (5/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 29/04/2025

Compared to traditional speech features like MFCC or Mel Spectrograms, self-supervised features are much superior in capturing allophony. (4/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 29/04/2025

In short, yes! Even though self-supervised speech models are trained only from raw speech, they cluster via allophonic variations, i.e., different surrounding phonetic environments. (2/n)

100 0

Kwanghee Choi @juice500ml.bsky.social · 29/04/2025

Can self-supervised models 🤖 understand allophony 🗣? Excited to share my new #NAACL2025 paper: Leveraging Allophony in Self-Supervised Speech Models for Atypical Pronunciation Assessment arxiv.org/abs/2502.07029 (1/n)

21510 0