A Brainstorm on Crowd-Sourcing Documentary Film Logging and Transcription…
The below is excerpted from an email I sent to some friends at an unnamed tech company, regarding my documentary This is Fi. I think this would actually work as a way to crowdsource documentary film logging and transcription – allowing people to donate labor instead of cash.
The Problem:
* We have over 100+ hours of footage (video diaries) of a man who speaks accented English.
* The first step in turning those 100+ hours into a 90-minute movie and a rich media website is transcribing and logging the footage.
* The cheapest cost I’ve found for this task is $50/hour of footage by an Indian offshoring company called
REDACTED. This would mean a minimum of $5000 to get started on the movie.* If we don’t hire it out, we’re looking at 400-600 man hours of doing the logging/transcription ourselves.
We were just granted non-profit sponsorship for our doc… Which means that my partner and I are about to go beat the bushes for $5K in donations just to transcribe the film. Ouch!
Buy what if we weren’t asking for cash? What if we could ask – even our brokest friends – for a donation of time and boredom.
Proposed Solution:
* What if we broke our hundred hours of footage into manageable sizes (15 seconds) and crowd-sourced the transcription. (Like Aaron Koblin’s work with Mechanical Turk.)
* The structure would be the video version of the recaptcha project – people’s transcription would be checked against other transcriptions… Sometimes a user would be providing a first pass, often they’d be validating.
* “Donations” of transcription would be charted against the goal (think “Kickstarter”). People would be able to see the impact they’d had on the process.
Why this matters:
* Distributed logging could be a significant cost-cutter not only for doc projects, but for corporate video/reality shows/etc. It’s a way to aggregate the time of interns/support staff/even audience members into meaningful work without wasting resources on managing volunteers.
* Metadata collected here could be an early part of the feedback loop – just by adding opinion information to the logging data… For example: what did you see? How compelling was it on a scale of 1 to 5?
* Even if not using volunteers – paying micropayments of 10 cents per 15 seconds would be a drastic reduction in cost over conventional logging. Paying by the second could bring a huge savings to production co’s IF there was some way to track the data from the original source footage, to the review area, and back.
The question:
* Is this possible?
* What are the stumbling blocks of breaking-down and reconstituting the footage and associated metadata? Can editing software talk to the net?
* Do you know of a solution that does this already? This is probably ridiculous – but could you batch-upload to YouTube and then scrape YouTube comments for logging info?
What do YOU think? None of us have money to donate to a film… What about time? A little effort? What if it was a game? Would you compete against friends? Would this be more/less boring than minesweeper/solitaire?
Did it work? I’m curious about it let me know
crazy idea guys
We didn’t do it… Currently working with Metadata tools from Adobe and manual (old fashioned) logging/transcription.
Crowd Sourcing In The Human Cloud:
I too had this idea and recently did a Google search to see who else was thinking like me.
I most definitely think that a “free service” like youtube could be used for storing the video. These could be marked as “Private” or “friends” only and embedded on your site (depending on your distribution model).
I would try to use a seperate site running some form of SQL for tracking Video with metadata. Use a link to the youtube videos. You could make metadata, tag cloud, etc. visible to users so that they could “drag + drop” things they feel are related and keeping a standard naming/description. It would be extremely easy to pull this data later as a user comment would have the video ID number and sql string data to parse and compare with others of the same video ID and translate to whatever format you want for whatever program. You could even make a system to mark groups of video, so you could combine the smaller clips and using info during the batch video splitting process recombine with timecode/tape info. The uploading process is where the problem will probably occur. Research into the youtube(or other video sites) API would give immediate information if this is possible. Original CPU time for Compression of files for upload would be enormous.
Another thought would be allowing “translations” to be preformed at the same time. Many people speak a second language, why not leverage this knowledge as well. Instant market increase. We could share our knowledge and story with even more people, without being limited by the financial marketability of the piece or long grant applications we never finished
A great opportunity would be to promote something like this at SXSW Interactive, lots of eager people there. I currently have a 1 TB full of Documentary footage that I haven’t made progress on because no logging was done at capture, I let the tape run while I did other work.
Let me know your thoughts. My email should be listed for you with this post. Sometime in March I would LOVE to start some coding to make this possible for myself and others. Let’s collaborate.
We need more documentaries in this world. We are a great machine, capable of many glorious things. Let us get to work.
Patrick,
If you know how to do this – I’m 100% game to help… Just not sure how much help I can be. This is waaaay beyond my coding skills, but I’m happy to pimp the results of your coding skills, help wrangle people, assets, etc.
I’ve been working with Adobe as a beta tester for some of the new metadata features they’ve got coming in CS5 which seem like they’d help – immensely – once you’ve got everything logged and transcribed. I haven’t seen this work yet, but I’m optimistic and hope that this would close the loop eventually (like in your example of simultaneous translation). I’m mainly a FinalCut user, but the metadata functionality is enough to woo me out of my comfort zone – esp. since my doc is a cuts-only affair.
What are your thoughts on this? What would it take to get started? How do you make your SQL database talk to YouTube?
Best,
David