Why socialists can’t do basic math

Dubai is no longer a non-extradition mecca, which is a good news!

https://meduza.io/news/2024/03/13/futbolista-spartaka-kvinsi-promesa-arestovali-v-dubae-po-zaprosu-prokuratury-niderlandov

Hint if you are voting today - use your own pen, it is not uncommon to use disapearring ink in provided pens.

Французские войска могут присутствовать в Украине. Никаких красных линий не будет. Я, как президент Франции, принимаю это решение. Мы делали все возможное для дипломатического урегулирования, но с Россией нет смысла вести переговоры. Украина должна одержать победу.

Макрон, в интервью France 2

gm

Meanwhile this is a stats of voting around my house

New toy, really impressed how smooth modern assist works and how nice is continuously variable transmission could be on the bicycle.

Meanwhile USA surpases G7 in GPTGDP growth

✨ Supervoice

I have published results of last two months of work, the Supervoice - natural sounded voice synthesis using modern SOTA architectures, trained from scratch. It is beta because: it doesn't work for large sentences, it doesn't support number literals (you have to write them using letters) and it works only with english (but i have explored options of training other languages and nothing really stops to do so, except less data). Almost anyone can run it anywhere where you can install pytorch and torchaudio (no other dependencies).

https://github.com/ex3ndr/supervoice

Я думал будет застой в политической жизни, а путин опять взрывает дома.

В один день:
- Письков говорит что россия воюет с украиной (впервые)
- Через 5 часов появляются первые уведомления что начали рассылать электронные повестки в москве
- Вечером (москвы) терракт

Ну да ну да, чистая случайность.

Кстати заходите в наш чатик - https://t.me/founders_chat

Remember there were app Amo, by Zenly founders?

gm

Dataset with phoneme-aligned librilight datasets for voice training at scale.

Dataset has almost 60k hours of english audio in 2TB Tar file, which i have processed with whisper and then aligned with MFA.

I am planning to train a base model for voice processing using this dataset and then finetune it to provide high quality voice, diarization and voice enhancement models. May be you will find it useful too!

https://github.com/ex3ndr/supervoice-dataset/

Recently i tried to login to my Heroku account and it wasn't working. I was paying like $100/mo for zero users. No notification, no explanation, account is still running i just doesn't have an access to it.

God i hate clouds.