Lightnews — Scholar-powered news

Christoph Minixhofer @cdminix.bsky.social · 20d

TTSDS2 is one of the papers accepted by the @neuripsconf.bsky.social area chairs but but rejected by the senior area chairs with no explanation as to why. A bit frustrating after the long review process.

Christoph Minixhofer @cdminix.bsky.social · Aug 29

100% agreed, also crisps are snack, not a side dish for lunch

1

Christoph Minixhofer @cdminix.bsky.social · Aug 24

Accents are also best seen as a distribution, not a group of labels imo. We tried to incorporate some proxy of accent in TTSDS2, but a simple phone distribution did not work all that well, probably because it’s hard to disentangle from lexical content…

Nina Markl 🏳️‍⚧️🏳️‍🌈 @ninamarkl.bsky.social · Aug 22

in honour of interspeech this week, i’d like to issue a reminder that everyone has an accent, and that’s beautiful, actually: www.isca-archive.org/interspeech_...

www.isca-archive.org/interspeech_...

1 1

Christoph Minixhofer @cdminix.bsky.social · Aug 21

It's been a great #interspeech2025!
I presented a TTS-for-ASR paper:
www.isca-archive.org/interspeech_...
And one on prosody reps: www.isca-archive.org/interspeech_...
There were many interesting questions & comments - if you have more and didn't get the chance feel free to send me a message.

2

Christoph Minixhofer @cdminix.bsky.social · Aug 20

I’ll will be presenting this tomorrow at 8.50 at #interspeech2025, come by if you’re interested in prosodic representations!

Christoph Minixhofer @cdminix.bsky.social · Jun 12

🧪 SSL (self-supervised learning) models can produce very useful speech representations, but what if we limit their input to prosodic correlates (Pitch, Energy, Voice Activity)? Sarenne Wallbridge and I explored what these representations do (and don’t) encode: arxiv.org/abs/2506.02584 1/2

Prosodic Structure Beyond Lexical Content: A Study of Self-Supervised Learning

People exploit the predictability of lexical structures during text comprehension. Though predictable structure is also present in speech, the degree to which prosody, e.g. intonation, tempo, and loud...

arxiv.org

1

Christoph Minixhofer @cdminix.bsky.social · Aug 20

Thank you to everyone who stopped by, I’m grateful for all the feedback and interesting questions #interspeech2025

1

Christoph Minixhofer @cdminix.bsky.social · Aug 19

In other news — if you’re an early bird and at #interspeech, feel free to drop by my poster presentation on scaling synthetic data tomorrow - who doesn’t want to chat about neural scaling laws early in the morning!
App: interspeech.app.link?event=687602...
Paper: www.isca-archive.org/interspeech_...

1 2

Christoph Minixhofer @cdminix.bsky.social · Aug 19

I tried: “what sport should I pick up?” and for my original (male) voice it responded with “association football is the most popular sport in the UK”. For my female one… “oh, for a newbie? Something easy like […]” — Goes without saying that research into these biases is important. 2/2

1 3

Christoph Minixhofer @cdminix.bsky.social · Aug 19

A highlight at #interspeech so far: the “hear me out” show&tell in which you can check how the spoken language model Moshi responds based on if it’s your voice or a voice converted version to the opposite gender.
Check it out here shreeharsha-bs.github.io/Hear-Me-Out/
1/2

Hear Me Out

Interactive evaluation and bias discovery platform for speech-to-speech conversational AI

shreeharsha-bs.github.io

1 1 3

Christoph Minixhofer @cdminix.bsky.social · Aug 18

If you’re interested in ASR for low resource languages, come by at 14.30 in Poster Area 09 at #interspeech today! I’ll be presenting this paper by Ondrej Klejch et al. arxiv.org/abs/2506.04915

A Practitioner's Guide to Building ASR Models for Low-Resource Languages: A Case Study on Scottish Gaelic

An effective approach to the development of ASR systems for low-resource languages is to fine-tune an existing multilingual end-to-end model. When the original model has been trained on large quantiti...

arxiv.org

2

Christoph Minixhofer @cdminix.bsky.social · Aug 11

Looking forward to present a bunch of things at #INTERSPEECH and #SSW - will put the details here once my thesis final draft is done, which will probably be on the plane to Rotterdam.

Christoph Minixhofer @cdminix.bsky.social · Jul 4

One day until the Q2 ttsdsbenchmark.com update. We‘ll see which TTS system tops the leaderboard this time - some new ones have been added that could shake things up.

Christoph Minixhofer @cdminix.bsky.social · Jul 3

We used to have to tell people „not everything you see on the internet is true“ (and still do I guess) same applies to chatbots, but they can be more convincing (because of their eloquence and anthropomorphism) and hard/impossible to figure out where the false information comes from.

3

Christoph Minixhofer @cdminix.bsky.social · Jul 2

Followed your advice and can confirm “Ughaaaghaghaa” was my reaction as well.

Erika Ishii @erikaishii.bsky.social · Jun 28

You ever watch a film and just know it’s a seminal medium-defining work of peak interdisciplinary storytelling all you can say in the moment is “Ughaaaghaghaa” and then cry?

So yeah everyone watch K-Pop Demon Hunters.

Christoph Minixhofer @cdminix.bsky.social · Jun 30

This figure motivated a lot of my PhD (or at least nudged me into a direction) -- check out arxiv.org/abs/2110.11479 (Hu et al.) if you haven't come across it before, it really frames the problem of synthetic/real speech distributions well.

Figure showing two overlapping bell curves representing data distributions. The green curve on the left is labeled ‘synthetic data distribution’, and the black curve on the right is labeled ‘true data distribution’. The horizontal axis is divided into four regions: ‘artifacts’ (only covered by the green curve), ‘over-sampled’ (where the synthetic curve is higher than true), ‘under-sampled’ (where the true curve is higher than synthetic), and ‘missing samples’ (only covered by the black curve). Caption: Fig. 1 describes the gap between synthetic and true data distributions partitioned into four regions.

Christoph Minixhofer @cdminix.bsky.social · Jun 29

Spotted a Norwegian flag across the Firth of Forth, didn’t know Norwegians had hytte on this side of the North Sea as well!

Norwegian flag in a sunny and green scene in Scotland with water and a bridge in the background.

Christoph Minixhofer @cdminix.bsky.social · Jun 27

More details on this soon! Also this weekend is the last chance to submit your TTS system for the next round of evaluation (Q2 2025) by either messaging me at [email protected] or requesting a model here: huggingface.co/spaces/ttsds...

arXiv cs.SD Sound @cssd-bot.bsky.social · Jun 25

Christoph Minixhofer, Ondrej Klejch, Peter Bell: TTSDS2: Resources and Benchmark for Evaluating Human-Quality Text to Speech Systems https://arxiv.org/abs/2506.19441 https://arxiv.org/pdf/2506.19441 https://arxiv.org/html/2506.19441

1 1

Christoph Minixhofer @cdminix.bsky.social · Jun 27

It’s amazing how a days work can stretch out over a fortnight, and a week of work can be compressed into 24 hours sometimes…

Christoph Minixhofer @cdminix.bsky.social · Jun 25

I wonder if there are naturally left-curling and right-curling cats, or if all cats curl both ways.

Unseen Japan @unseenjapan.com · Jun 25

Not Japan-related, but since we all need a distraction from The Horrors, Takaya Suzuki points out a study that examined 408 sleeping cats and found the majority (65%) curl leftwards.

I'm not sure how useful this information is, but...it's yours now.

Post by suzuki_takaya

衝撃の事実。ネコは１日12−16時間寝るけれども、この丸まったポジションの左巻き・右巻きのどちらが多いか。YouTubeで何匹もの眠るネコを観察し、266匹（65％）のネコが左巻き、142匹のネコが右巻きに寝ていたことを発見。有意な差が見られたとのこと。人類史上に残る幸せな研究といえる。

1

Christoph Minixhofer @cdminix.bsky.social · Jun 23

I’ve only really encountered people trying to avoid sounding like AI… but it makes sense that it would alter how people speak if they interact with it a lot. Makes me sad though since it pushes people towards the mean, which is always the most boring.

Eryk Salvaggio @eryk.bsky.social · Jun 22

It’s funny that “nuance” is one of the words LLMs tend to avoid www.theverge.com/openai/68674...

You sound like ChatGPT

AI isn’t just impacting how we write — it’s changing how we speak and interact with others. And there’s only more to come.

www.theverge.com

Christoph Minixhofer @cdminix.bsky.social · Jun 23

This made me think back to the last couple books I read… My last one that I could actually describe as wholesome is 8 books ago: A Wizard of Earthsea by Urusla Le Guin. Otherwise, anything by Terry Pratchett?

1

Christoph Minixhofer @cdminix.bsky.social · Jun 18

Mileage may vary based on how long you and your poster wait at a bus stop in Edinburgh city centre and the current foot traffic and the number of American tourists with family members in your academic field passing by.

Christoph Minixhofer @cdminix.bsky.social · Jun 18

Pro tip #1: don’t use a poster tube when travelling to and a from conferences, people might come up to you and ask about your research. Pro tip #2: get a poster tube if you don’t want to talk about your research anymore once you’re done with the conference.

1

Christoph Minixhofer @cdminix.bsky.social · Jun 17

This is from „Text-to-Speech Synthesis“ (2009) - and honestly, I like a bold claim like this (even if wrong), and who knows what I would’ve thought if I had done research at the time - and I wonder how many of my current beliefs will turn out to be wrong in ~15 years!

Christoph Minixhofer @cdminix.bsky.social · Jun 17

Taylor 2009:
„Sometimes the question is raised as to whether we really want a TTS system to sound like a human at all.“

me: I wonder where this is going

later:
„no matter how good a system is, it will rarely be mistaken for a real person, and we believe this concern can be ignored.“

me: oh no

1