These are chat archives for yaskyj/fastcaption

6th
Feb 2015
Justin Rogers
@yaskyj
Feb 06 2015 20:01
They're getting ahead of us!
Justin Rogers
@yaskyj
Feb 06 2015 20:11
Unfortunately, I will not be able to make the meeting tonight. Is there any other time that you guys have available during this weekend?
Justin Rogers
@yaskyj
Feb 06 2015 20:22
@MichaelLockrey I also had a question about the "improve creative aspects" card. I'm not familiar Xero, but what how do you envision the collaboration via Twitter/Facebook?
Just a tweet or facebook message news feed entry giving the link and a small note about it?
I'm guessing that will require some kind of balance between the ability to make transcripts private so that they can't not be defaced and collaboration.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 20:37
Hi guys - I can be available most mornings early - if that suits (including the weekends)
Early is 5am Sydney time
Which is 10am West coast USA time
Or alternatively, afternoons here on Sunday (which will be Saturday evening over there)
In relation to Twitter I was just thinking that you could handle a transcript in a micro-task manner
And disseminate it amongst your followers, or a private pre-screened crowd of supporters initially etc
Michael Lockrey
@MichaelLockrey
Feb 06 2015 20:42
Then each tweet could comprise one caption block or chunk (in 140 characters)...
You could also get multiple responses for the same block of captioning to improve quality of final product
Alternatively it could be as simple as a Google Docs interface
With each response showing up in a segmented / colour coded Google Doc
So craptions that haven't been edited could be red, edits made could be green and QA flags could be orange (where there's obvious spelling mistakes typos etc)
The whole strategy behind this is that it's hard work to do transcripts - so why don't we break it down into smaller bite-sized chunks
Michael Lockrey
@MichaelLockrey
Feb 06 2015 20:48
For the crowd to re-assemble?
As you know one thing I do at the moment is download a small MP3 audio file of the YouTube video
Which has the smallest possible file footprint etc
Would it be possible to then "chunk" the MP3 audio into the same bite sized chunks so they would play in each micro-task?
Eg 1-3 seconds, caption block one - and the related audio clipping, 3-5 seconds - caption block two and the related audio clipping plays etc etc
Justin Rogers
@yaskyj
Feb 06 2015 21:12
So in addition to the ASR file, the site also gets an MP3 file containing all of the audio?
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:25
No sorry I should have explained myself
When I do captioning jobs that do not have an ASR track
i.e. starting from scratch
I sync the transcript I create in oTranscribe et al with Sync Words
using the mp3 file as the media source
that Sync Words use to perform the syncing process
Justin Rogers
@yaskyj
Feb 06 2015 21:30
What do you use to rip the MP3 from YouTube?
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:31
I just use Clip Grab
It's very much a manual process at the moment
But it means that a standard YouTube file might be 10-20Mb in video formats such as mp4
But the related mp3 might only be 1/2Mb or something less
Justin Rogers
@yaskyj
Feb 06 2015 21:33
Understood. I've been trying to see if there's some free API for ASR.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:34
Sync Words is the best I've found
You're welcome to have a look at Speechmatics too
It has a free trial with some initial credits
But it was very slow when I used it
I think YouTube's ASR time syncing is actually quite good - it certainly gets us 80% accuracy with the timing most of the time
The accuracy on the other hand is still very much not workable
The biggest problem with YouTube's ASR is that quite often syncs "orphan" single words
as a caption block
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:39
and they might only have .2 of a second duration assigned to them
Which means that the word flashes up and is gone much too quickly for us to decipher it when using the captioning
Justin Rogers
@yaskyj
Feb 06 2015 21:39
When I looked at syncwords it said that it charges for anything over two minutes?
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:40
Yes - I think that is correct
But we can try it for shorter videos?
The other option is to have a YouTube account that we use for syncing purposes?
But this would mean taking copies of media and uploading them to this channel
Justin Rogers
@yaskyj
Feb 06 2015 21:41
We can definitely see about using it for shorter videos.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:41
along with the completed transcripts
Justin Rogers
@yaskyj
Feb 06 2015 21:42
Arthur and I were just searching for something that would cover all/more use cases.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:42
I think that would take too long and we'd potentially run into copyright barriers / hurdles
Yes we need to be able to handle transcripts from scratch where there's no ASR
or being able to upload a transcript (BYO)
I have a good feeling about micro tasking the transcript somehow
and taking the pain out of it
If we could generate good quality captions we could pay for a Sync Words back end then
Justin Rogers
@yaskyj
Feb 06 2015 21:47
Breaking it down does seem like a good idea. Breaking it down would bea low barrier for people. I also had an idea about adding something that would take the most popular videos of the day/week/whichever to send out for people to transcribe.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:48
Yes that's the sort of stuff that needs to happen
But it needs to happen quickly and in real time
That's why I thought of Twitter
It's so fast
And everyone's got 140 characters in them!
;)
Michael Lockrey
@MichaelLockrey
Feb 06 2015 21:53
You can see that it's "hard work"
Now that you've had a chance to try and transcribe a few videos
Somehow we need to make this less painful
The big problem with Amara is that they do have a captions requested team
where you can post up a video and request that it be captioned
But they usually sit there for weeks until one of the regular volunteers rolls their sleeves up and does them on their own!
Justin Rogers
@yaskyj
Feb 06 2015 22:00
I've been looking through Amara's python code and trying to reverse engineer some things. They create a separate class for each type of video, so we'll have to look into the different APIs besides YouTube for Vimeo, Yahoo, etc.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 22:00
Right
Facebook video is huge now too
That's interesting how they create a separate "class" for each video hosting platform
But not a total surprise I guess
Justin Rogers
@yaskyj
Feb 06 2015 22:02
But once we're done with the YouTube api, I don't think the other APIs will be that different as long as their serving HTML5. We'll be able to use the same frontend code for the most part.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 22:02
Yes
I'm very hopeful that the move to HTML 5 will help us
Justin Rogers
@yaskyj
Feb 06 2015 22:02
Yeah, definitely.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 22:02
Does YouTube offer a streaming mp3 version?
You know how I get the streamed HTML 5 video
Do they do the same with video?
Justin Rogers
@yaskyj
Feb 06 2015 22:03
I haven't seen one.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 22:03
OK
Justin Rogers
@yaskyj
Feb 06 2015 22:03
It's almost like you get everything or nothing.
If you follow their terms of service.
Michael Lockrey
@MichaelLockrey
Feb 06 2015 23:05
I'm just chatting to Sam Cartsos at Sync Words right now on Skype
He told me that they extract the audio from the media files themselves to do their syncing processes
As I asked him whether uploading a low quality mp3 audio file would hinder / detract from the syncing process they do