Also, I have that tested (auto download) with YouTube-Dl, it works fine but haven't put it into the frontend, but may as well, it helps a lot on your own instance so you don't have to download it first and then upload it
Right now the limit is 100MB because of Cloudflare, no length limit, costs $1.30/h to run this, that's enough for a 2x RTX A6000 on Vast.ai can you check out the specs there
I've got people to test and usually with the 'medium' and 'large' models it works really efficiently. Honestly, I just use the large model for everything, because might as well have the best quality if you're going to do the effort.
Well the code is open source, I don't know what I plan to do with this, depends on how people like it, but for the meantime you can use it to transcribe stuff for free which is a victory unto itself
This was just a personal project a couple hours ago so it's not setup properly to do safe reboots and a lot of other things, I was just using it locally and now it's in in the wild, will take some time to get everything refined and professional
I just rebooted the server, now if the websocket disconnects (person closes the browser) it will kill the processing and move the queue, so that should help unclog the queue. I'm going to add a couple more queue touchups and then it should be stable again (no reboots), but running well in the meantime
Yeah I apologize, the queue is a little messed up. It's not showing your progress properly and it's not stopping people's processing if they left (their websocket dies). I'm going to fix these and reboot the server and the experience will be a lot better, sorry. It is working properly though and transcribing all this stuff but the queue needs some TLC, brb.
Yeah someone submitted a PR for those to be fixed, I'm just wary about restarting the server because I haven't setup a way to be able to reboot without losing the websockets
Well those inputs should all error unless they are a valid value.
Yes if nothing is input it will automatically detect the language based on the first 30s of input
Right on, looking forward to it! Yeah I saw that module and was planning to use it but I just wrote up an Express/Node implementation first and never really looked back. But looking forward to collabing I will await your issues, cheers!
Thanks! I wrote it for myself over a weekend and have really enjoyed it ever since I'm glad others were able to get something out of it! It seems to run pretty well but I have some improvements planned, first is I will take the whisper output and feed it to you when you're in the queue so you can see them progressing. Will be pretty trivial to implement but I am feeling bored in the queue at the site atm so that is the next killer feature lol
Yeah I was paying $100/month for transcription services and turns out Whisper with the large model was much more accurate, and I didn't like the UI, I much prefer to just use this app as opposed to the paid service, and I chose it because it was the cheapest by far ($100/30h) as opposed to most of the other paid services which were $10 an hour which to me was a bit much really. But Whisper is really a game changer I don't know how those companies stay in business really.
Whisper does pretty well, even with background music and things like that, I think you're working with a pretty weird subsection of recorded audio that won't work, for that edge case to work you'll very likely need to train your own model.
It sounds pretty good, this is my first time hearing about it but it looks good. Even if it does detect that they are separate entities talking, how does it label it in a way that's helpful/useful for you? I guess it comes out as 'Speaker 1', 'Speaker 2', etc in the end? And you can find/replace the speakers with the actual people?
Hey, I know the feeling, I felt bad when I had my GPU just sitting there and it's just a little Vast server lol. If you want to use your hardware to run this software I'd be more than happy to help get it setup!
I'm just running this off of a 2x RTX A6000 server on Vast.ai at the moment, about $1.30/h and then using nginx on another server to reverse proxy it to Vast
Yeah it's all ready to go using LibreTranslate, they have about 25 languages, maybe I'll finish that this weekend and put it up, it's really inexpensive to make the translations compared to making the original transcription so may as well. Coming soon!
No problem! I am just seeing how it runs, I might throw up a referral link to Vast and put up a tutorial on how to host your own service, maybe that can offset the cost a bit? The current server is $700/month, maybe it could just run off donations who knows
Yeah Whisper is top of the line, they posted their performance compared to the industry standard and it's right in there at the top.
It should show the data via processing.. it's setup to just take whatever stdout/stderr comes back from Whisper and send it directly to the frontend via websockets, I'm surprised you got stuck there :thinking:
Hey, glad I could be of use. The problem with Whisper is that it needs a lot of GPU. Actually my Mac can't even use my GPU so right away I had to get it up on a server, but Whisper is so powerful and it's so amazing that it's open source I am surprised nobody did this yet. I could see them charging for it but may as well use it anyways, the other services are insanely expensive ($10/h?!) and I don't really like their UIs to boot lol
Yeah it's a hard case, Whisper with the large model is among the cutting edge in the business so if the static is bad and the quality is low there's not much you can do but wait for better AI, or fix whatever they get wrong by hand, but Whisper AI is on the cutting edge so you might have to wait for a bit lol
Yeah it does, I think there's a bug with it for saying what your position is, but when the others are done it will start correctly even if the frontend shows like position -2 or something. There's 2 uploads in the queue atm so not bad