Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: I made a free transcription service powered by Whisper AI

freesubtitles.ai

Post reply on HN
Print view View original

91–100 of 135 posts

Re: Show HN: I made a free transcription service powered by Whisper AI

#91

Post by ivalm »

kkielhofner wrote: post #55
jcalvinowens wrote: post #52

Earlier quoted context omitted.

> the sampling, the jargon, the 10 codes, the background noise, static on analog systems/drop outs on digital systems, rate of speech, etc Is the reduced bandwidth really the most significant problem? Naively I'd think everything else you mentioned would matter a lot more, I'm curious how much you experimented with that specifically.

When it all comes together it's kind of a nightmare for an ASR model. There were plenty of times in reviewing the recordings and ASR output where I'd listen to the audio and have no idea what they said. I'm not sure which contributes most but I know from my prior experiences with ASR for telephony even clean speech on pristine connections does much worse with models trained on 16 kHz being fed native 8 kHz audio that…

Could one train an interpolation layer (eg take a bunch of 16k audio, down sample to 8k, train 8k->16k upsampler)? Or better yet (but more expensive), take whisper, freeze it, and train the upsampler on whisper’s loss.

Re: Show HN: I made a free transcription service powered by Whisper AI

#92

Post by kkielhofner »

ivalm wrote: post #91
kkielhofner wrote: post #55

Earlier quoted context omitted.

When it all comes together it's kind of a nightmare for an ASR model. There were plenty of times in reviewing the recordings and ASR output where I'd listen to the audio and have no idea what they said. I'm not sure which contributes most but I know from my prior experiences with ASR for telephony even clean speech on pristine connections does much worse with models trained on 16 kHz being fed native 8 kHz audio that…

Could one train an interpolation layer (eg take a bunch of 16k audio, down sample to 8k, train 8k->16k upsampler)? Or better yet (but more expensive), take whisper, freeze it, and train the upsampler on whisper’s loss.

"I understand some of those words."

Hah, in all seriousness I'm more of a practitioner in this space. If this was something I absolutely needed to get done who knows where it would have went. For a little side hacking project once I encountered these issues I moved on - back in the day expectations were lower for telephony and the 8 kHz aspire models and kaldi were adequate to get that "real work" done.

Re: Show HN: I made a free transcription service powered by Whisper AI

#93

Post by jazzyjackson »

TheBrokenRail wrote: post #70
mayeaux wrote: post #65

Earlier quoted context omitted.

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

> I'm just wary about restarting the server because I haven't setup a way to be able to reboot without losing the websockets Wait what. Not being able to safely restart the server sounds like a disaster waiting to happen.

> losing the websockets

users would lose the session and have to start over, not the end of the world

Re: Show HN: I made a free transcription service powered by Whisper AI

#94

Post by jazzyjackson »

bergenty wrote: post #89
mayeaux wrote: post #81

Earlier quoted context omitted.

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

Isn’t this very expensive to host? Are you aware this could cost A LOT?

in another comment they state:

> I'm just running this off of a 2x RTX A6000 server on Vast.ai at the moment, about $1.30/h

whether that's a lot is a matter of perspective

Re: Show HN: I made a free transcription service powered by Whisper AI

#95

Post by n8cpdx »

This is a cool project. I’ve been very happy with whisper as an alternative to otter; it works better and solves real problems for me.

I feel compelled to point out whisper.cpp. It may be cheaper for the author but is relevant for others.

I was running whisper on a gtx 1070 to get decent performance; it was terribly slow on M1 Mac. Whisper.cpp has comparable performance to the 1070 while running on M1 CPU. It is easy to build and run and well documented.

https://github.com/ggerganov/whisper.cpp

I hope this doesn’t come off the wrong way, I love this project and I’m glad to see the technology democratized. Easily accessible high-quality transcription will be a game changer for many people and organizations.

Re: Show HN: I made a free transcription service powered by Whisper AI

#96

Post by oefrha »

kkielhofner wrote: post #66
mayeaux wrote: post #17

Earlier quoted context omitted.

I setup the server to only transcribe two files at a time, so yeah someone could abuse it for sure with two big uploads and stick everyone else on the queue. But for me, even a 3 hour video translates with large model in about ~30 minutes so it wouldn't be too bad, but hopefully everyone is conscious to not do that, so far nobody has abused it which is cool.

Me again - why two at a time? In my initial testing with whisper-asr-webservice and my RTX 3090 I could pretty easily throw ~10 different files at it simultaneously as there is some natural staggering between API entry, CPU conversion/resampling/transcoding of audio, the actual audio length, network effects like upload speed, etc. I also implemented some anti-abuse-ish features between traefik and Cloudflare that sho…

> I am just paying for a somewhat expensive server and I love how it's really fast but also I have a lot of free GPU time so might as well let others use it too lol.

They are donating some spare capacity.

Re: Show HN: I made a free transcription service powered by Whisper AI

#97

Post by JanSt »

seligman99 wrote: post #74

It broke when I tried to feed it an entire podcast file, but still, I took this as a push to try out Whisper AI for myself, turns out it's easier to use than I thought. Long story short, I used it to transcriptify a podcast: https://scotts-podcasts.s3.amazonaws.com/temp/whisper/Intern... Not sure if there's a use for this that's not me, but I like the idea of having subtitles for a podcast I'm listening to.

What tool did you use for the player-text presentation?

Re: Show HN: I made a free transcription service powered by Whisper AI

#100

Post by bayramn »

mayeaux wrote: post #35
sali0 wrote: post #29

Earlier quoted context omitted.

Very generous of you. I made a similar free service 3 years ago using much worse tools and it's so cool to see whisper making it all so much better and more efficient. Thanks for releasing for free

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

$700/month? Where digital ocean? I am new to python and ML, curious to know why..
Post reply on HN