@RSvoboda432@FFmpeg@Remotion Oh interesting problem. Sounds like some OCR would be involved? Tough to do efficiently with all those frames, especially if sometimes the words are actually "in progress"... 🤔
@RSvoboda432@FFmpeg@Remotion We can transcode audio segments in parallel, then join them at the end without transcoding. It removes a bottleneck where, before, AAC transcoding had to happen for the entire file in one big chunk. There's a good post on it on the Remotion Blog: https://t.co/CqFzGXuCDs.
@RSvoboda432@FFmpeg We're using it at Wistia for fast mp4 exports from our Editor and fast mp4 transcoding generally. Soon to be using it for on-demand HLS segment transcoding too. It's also in use by @Remotion. For long videos, it saves a lot of time!
A couple years ago, at a park in Austin, TX, I figured out how to seamlessly split and stitch AAC audio streams using just ffmpeg. It's been in production use at Wistia for over a year now. Friday I released a public repo that demonstrates how it's done! https://t.co/FEUimTNGlG
@JNYBGR Glad to help and excited to see this technique make its way into Remotion!
As you can imagine, I spent many days figuring this out, and, while the technique was super obscure, maybe the hardest part was everywhere on the internet telling me it was impossible.
It is permissible in English for a preposition to be what you end a sentence with.
The idea that it should be avoided came from writers who were trying to align the language with Latin, but there is no reason to suggest ending a sentence with a preposition is wrong.
I was so psyched when I got it because all my research on the internet up to that point had been saying it's not possible, that you have to transcode it at the end. But it just didn't seem right to me! So I read docs and experimented until I worked out the math. And it worked! 😀
A couple years ago, at a park in Austin, TX, I figured out how to seamlessly split and stitch AAC audio streams using just ffmpeg. It's been in production use at Wistia for over a year now. Friday I released a public repo that demonstrates how it's done! https://t.co/FEUimTNGlG
This technique can be used to avoid extra latency to combine audio when doing parallel rendering of audio or video. A variation of it can also be used for rendering and delivering individual HLS segments on demand.
You cannot solve self driving by driving in a parking lot and you cannot solve robotics in a lab.
Much before we build physical AGI, we gotta force robots to leave the lab for the real world and collect lots of data about the nature of physics, reality and humans.