Peter

GitHub - pevers/parkiet: Parkiet is a 1.6B parameter Dutch text-to-speech model (TTS) Parkiet is a 1.6B parameter Dutch text-to-speech model (TTS) - GitHub - pevers/parkiet: Parkiet is a 1.6B parameter Dutch text-to-speech model (TTS)

github.com/pevers/parkiet for open weights and training code on Google Cloud TPUs using free research credit. #MLSky #MachineLearning #TTS #ML

Can you train a state of the art text-to-speech (TTS) model with little resources? It seems you can! I've used Google's free TPU research credit to train a Dutch version of "Dia" called Parkiet, a natural conversation speech model and the quality is on-par with ElevenLabs.

Any idea how to get a hold of Google's free Research TPU credits? I get these assigned for 30 days but there is either no capacity or the accelerator is not available at all #ml

Thank you! There are still many things I have in mind to improve, but there is a solid base to practice with immersion.

Hah I doubt it because they make mistakes

Three other people have now surpassed me in all-time XP earned on my own language learning app! Great milestone. #buildinpublic #languagelearning

Our Weekly Update #18 is live! 🎥✨

Watch now: https://youtu.be/G2FQRFcLawQ
Don’t forget to subscribe & hit the bell 🔔!

Subscribe to our weekly newsletter!
👉 https://urlroulette.net/newsletter/subscribeform 🚀

@pevers.bsky.social
#christmas #marketplace #onlinetv

Reached the front-page of HN and now the app is really hammered by traffic 😬.

Got my first sale for Fluentsubs! Key things to make it happen:

- Shipped an awesome feature for beginner learners
- Made some exercises public to check out w/o an account
- Locked a lot of the features behind a "Pro" paywall and made it clear what the benefits are

#buildinpublic

Finished the Duolingo-like daily exercises for news content. You can try it out at app.fluentsubs.com/exercises/da... for 11 languages (will add more soon). Tomorrow I'll finally do some proper marketing and fix the membership option (basically everything is free) #buildinpublic #langsky

Thanks I’ll check it out! Almost done implementing it

DuoLingo like exercises for YouTube content is coming to Fluentsubs! #buildinpublic #langsky

Thanks! I have many ideas for improvements but didn't yet manage to get them deployed.

Discover Fluentsubs on Uneed Turn YouTube videos into language learning lessons

I launched Fluentsubs on @uneed.best today at www.uneed.best/tool/fluents... #buildinpublic

Fluentsubs Transform your favorite videos—from trending shows to viral clips—into powerful language lessons that stick. Supports over 32 languages.

Example of Macrons speech about Europe tonight app.fluentsubs.com/stream?v=cm7...

Videos with excellent transcripts can now be publicly shared! #buildinpublic

In a major upgrade to Fluentsubs I've added "cloze" for subtitles. Based on the accuracy of the transcript and the word length, random gaps are created as a listening exercise to the user to bridge the gap between reading/listening skills in a language. #buildinpublic #langsky

I’m fiddling with the idea of just running tests locally and signing them automatically. Basically a decentralized CI/CD . User X signs that commit Y runs fine -> merge!

But how do you do test selection in a large monolith? How do you know which parts are affected?

Thanks! Luckily it was during the night so I couldn’t ship more bugs

Fluentsubs made it to the Hackernews "Show HN" front! news.ycombinator.com/item?id=4290...

Now I suddenly have the problem of not being able to YOLO deploy because I have users 😅.

I've also shipped better transcription models, more feedback and a fine-grained control adding new words to a deck!

Fluentsubs just got smarter! 🚀 Now you can pick only the part that you need for your deck—perfect for short spaced repetition cards. Check it out! 📚✨ #LanguageLearning #buildinpublic

Foundation models are overhyped. SAM2 was hard to fine-tune and we got some great results after tweaking the training process a lot. Decided to give YOLO11 another try (even after the crypto miner debacle) and it seems to be on par with SAM2 in terms of quality and much much faster!

Had some nice feedback from the first users and implemented the playlist feature, and added some basic channels so beginner language learners are not completely overwhelmed. The videos are pre-transcribed so that they can be watched immediately! #langsky #japanese #buildinpublic

Would be curious about this too

Thanks for trying! Will fix it!

Fluentsubs Transform your favorite videos—from trending shows to viral clips—into powerful language lessons that stick. Supports over 32 languages.

🚀 Just gave fluentsubs.com a complete makeover!
✅ New landing page – cleaner & easier to grasp
🎨 Fresh logo – friendlier & polished

I’d love to hear your thoughts! Put in a lot of effort to make everything super clear. #buildinpublic

I think it is possible to create a separate file and feed that to the context window using ‘@‘. But it is a bit of a hassle.

GitHub. Main reason it is being so used to it for the last 10 years. I’m also addicted to the GH trending feed to scout for new stuff

I noticed a huge difference with using Claude + Sonnet 3.5 instead. Less mistakes and way more context is taken into account.