Thank you Emmanuel! I am trying to make the animations and zooms better. If you have ideas to improve, please let me know. Sometimes I feel the screen is not moving properly but I can't put my finger on why exactly.

So DeepInfra works, but Meta-Llama-3.1-8B-Instruct-Turbo is much slower than Gemini Flash 8B or Groq Llama-3.1-8B-Instant. For the same transcript, DeepInfra took 27 seconds, Gemini 4 seconds and Groq 3 seconds. Costs are similar.

Thanks I’ll try DeepInfra Llama 3.1 8B Turbo. I noticed Llama supports a handful of languages, compared to over 100 languages for Gemini

How much is it on DeepInfra?

Yes it runs a small model called a MobileBert fine tuned on a token classification task (here: punctuation or not after this word?) it’s not perfect. I released a better version open source here that uses a LLM (Gemini Flash 8B) ldenoue.github.io/readabletran...

You can also read formatted transcripts with punctuation, paragraphs and chapters over at Scribe www.appblit.com/scribe

screenrun.app (my app)

Check screenrun.app for a web based solution that works anywhere

Check out my app if you like reading transcripts in addition to listening www.appblit.com/scribe

And here the full transcript with summary using the free Scribe (disclaimer: my app) www.appblit.com/watch?v=fRyy...