emacsconf-org
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Translating captions


From: Sacha Chua
Subject: Re: Translating captions
Date: Fri, 09 Dec 2022 09:33:08 -0500

Esteban Ordóñez <quiliro@riseup.net> writes:

Hello, Esteban, Jean-Christophe, all!

> The only thing that I do not agree with is that it takes every minute of
> talk 10 minutes of transcription.  I can do it in 2 minutes, including
> correction and review, even though I am not a professional transcriptor
> or native English speaker.
>> And even if Whisper is a bit slow on your machine (on that 8 cores
>> machine it used 50% of the processing power for a bit more than the
>> length of my presentation to complete the transcription) it is so
>> accurate that I would not want to spend too much time trying to do
>> that myself.
> I have a Core 2 Duo.  But will try on another machine, sometime.

I'm delighted that y'all want to get started already. <laugh> If there's
a talk that you particularly want to harvest the Q&A from and you don't
want to wait for our (probably not-very-optimized) setup of Whisper to
get around to it, please feel free to work on it either by listening to
it (instead of waiting for a transcript) or doing the speech-to-text
conversion on your machine.

JC, your experience sounds pretty neat! I always get tempted to upgrade
my laptop during EmacsConf... I've heard good things about
https://github.com/ggerganov/whisper.cpp (a C/C++ Whisper port) on Macs,
so you might get even better performance with that.

In general, since people tend to be busy and all of this stuff is a
nice-to-have, I try to optimize for volunteer energy/interest instead of
asking people to do stuff that a computer can do reasonably well. But
it's totally cool if you want to get started.

I have a few other tasks I need to finish since I put them off while
cramming things for EmacsConf, and then I can get back to all the fun
EmacsConf things. There are also some commercial speech-to-text services
that I might experiment with in case that helps us get captions together
for last-minute submissions, so I can test them with some of the Q&As
(speaker diarization might be handy) and figure out the workflows. I can
also look into some of the Python tools for doing that sort of stuff on
our own, so we can share the toolchain with other people.

So many opportunities to learn! =)

Sacha



reply via email to

[Prev in Thread] Current Thread [Next in Thread]