Yeah, it's the domain specific terms that really make or break these systems, especially in academic settings. These might suffice for business domains, but I've seen transcription quality drop quite fast for a STEM class or any non-Western humanities course. Ideally, there would be a feedback loop to these systems but I have yet to see one where you can send in corrections.
Katherine Deibel | PhD
Inclusion & Accessibility Librarian
Syracuse University Libraries
[log in to unmask]
222 Waverly Ave., Syracuse, NY 13244
From: Code for Libraries <[log in to unmask]> On Behalf Of Carol Kassel
Sent: Tuesday, February 12, 2019 4:42 PM
To: [log in to unmask]
Subject: Re: [CODE4LIB] [EXT] Re: [CODE4LIB] A/V and accessibility
Thank you so much for your replies! I'll reply to each of you individually as well.
In answer to your question about which auto-captioning solutions we're looking at, there are 2 main solutions we have our eye on. One is VerbIt and the other is Konch. Both appear to offer reasonable accuracy in the languages we need, though we are still evaluating. Still, as with any of these solutions, they miss some domain-specific vocabulary as well as anything that's mumbled or otherwise hard to understand. Also, we need to figure out our workflow for generating captions/transcripts, getting them into our infrastructure, and allowing for hand corrections and the workflow for revisions resulting from same. The devil is in the details!
> -----Original Message-----
> From: Code for Libraries <[log in to unmask]> On Behalf Of
> Carol Kassel
> Sent: Monday, February 11, 2019 11:31 AM
> To: [log in to unmask]
> Subject: [CODE4LIB] A/V and accessibility
> We're working on a roadmap for making A/V content from Special
> Collections accessible. For those of you who have been through this
> process, you know that one of the big-ticket items is captions and
> transcripts. In our exploration of options, we've found a couple of
> pretty good auto-captioning solutions. Their accuracy is about as good
> as what you'd get from performing OCR on scanned book pages, which
> libraries do all the time. One possibility is to perform
> auto-captioning on all items and then provide hand-captioning upon
> request for the specific items where a patron needs better captions.
> This idea will be better supported if we know what our peer
> institutions are doing... so what are you doing? Thanks to those to
> whom I've reached out personally; your information has helped
> tremendously. Now I'd like to find out from others how they've handled this issue.
> Thank you,
> Carol Kassel
> Senior Manager, Digital Library Infrastructure NYU Digital Library
> Technology Services [log in to unmask]
> (212) 992-9246
Senior Manager, Digital Library Infrastructure NYU Digital Library Technology Services [log in to unmask]