Machine captioning SRT testing
Re: Re: Re: Re: Re: : Re: Machine captioning SRT testing
2021-07-08 16:42:09
Ok, let me know which movies you enjoyed :)
About speech recognition, you're using a specific software? Which one?
Re: Re: Re: Re: Re: Re: : Re: Machine captioning SRT testing
2021-07-08 17:51:46
running google speech api. trying out rev.ai too. But quality differs. not sure where the issue is. i may be in way over my head over here.
It's really down to the transcription. was reading up on the documentation and essentially it's brute force ML where say based on sounds the person makes
eg
one to three
They will comb through their ML and identify the following at various confidence level
one two three
want two tree
want to trip
one to three
one to tree
So they will search through all the ML data they've been fed and realize the highest confidence would be
one to three
one two tree
so if their dataset has a higher occurrence of one two three then that's wat it recommends.
Using such an approach means only custom vocabulary can address. Eg certain phrases may be more commonly used in Hentai.
So the engine gets a bit smarter to check the custom set and gives it priority.
So you will have a hentai custom set.
Re: Machine captioning SRT testing
2021-07-09 09:14:55
Thanks, that's very interesting, and I didn't know about rev.ai.
Have you ever tried the Youtube auto-captioning tool (the one automatically run when we upload a video)?
If you did, can you tell me if the Speech-to-text api gives better result or is it the same?
Afaik, there is no drag'n'drop interface to use that API. Which programming language do you use?
Thanks!
Re: Re: Machine captioning SRT testing
2021-07-09 12:35:40
Oh check out the documentation
https://help.rev.ai/en/articles/3539014-examples-for-rev-ai-node-sdk-python-sdk-or-javascript
I'm running node
Re: Re: Re: Machine captioning SRT testing
2021-07-09 12:37:49
Rev has UI for u to upload Ur files btw so u dun really HV to code
Re: Re: Re: Re: Machine captioning SRT testing
2021-07-09 14:34:05
Thanks!
Re: Re: Re: Re: Re: Machine captioning SRT testing
2021-07-09 17:44:11
YouTube's pretty bad and jav ain't exactly viable to be hosted on YouTube. I tested used non sensitive video and it's terrible. Essentially there's 2 engine that speech to text uses to various degree of accuracy. One uses live streaming one used batch processing. YouTube based on my suspicion uses live translation hence the lower quality. Every video anyone uploads is scanned simultaneously for copyright and infringement which also includes translation before it's published.
Re: Re: Re: Re: Re: Re: Machine captioning SRT testing
2021-07-09 18:21:53
Ok, so I definitely have to try it to compare.
I'm not a developer, but I can do things in shell, PHP and Python, but I'm a crap with javascript.
Re: Machine captioning SRT testing
2021-07-10 02:39:34
If anything language is never a barrier. It's the desire and love of jav tat drives us.
Re: Re: Machine captioning SRT testing
2021-07-12 14:14:24
> Hey I've created a 3 transcription. possible to help me take a look?
From my point of view, Set A/D is the best. I think this one is the closest to the real meaning, but I may be wrong.
FYI, at the beginning of conversation, ただ今 means something like "I'm back home" and the polite answer is "おかえり" which mean "welcome back". (this is in case you wared why all the translation began by "right now").