Captioning Workflow // Joseph Polizzotto
As presented at the event
Brief description
This captioning workflow is an attempt to eliminate common pain points: generating a transcript, segmenting the transcript, and aligning the transcript with the audio.
We are interested in an open-source, accessible application that can integrate all of the steps in our existing workflow. See our "Needs for A Captioning Tool" document for more specifics.
Context
Our work is in disability services in Higher Ed (California Community Colleges and NYU, respectively). We are required to create captioned videos per section 508 of the Rehabilitiation Act. Many of colleagues are looking for methods to reduce cost and labor in the captioning process.
Stage of production:
We have created bash scripts that automate common steps in the captioning process. Script #1 is for getting a "raw transcript" from the YouTube auto-captions track (we use youtube-dl and Aeneas and Sed). Script #2 is for segmenting the edited transcript so that we have "caption-ready" chunks that conform to the DCMP's recommendations of quality captions. In Script #2, we also align the segmented transcript with the video via Aeneas, a Python library by Alberto Pettarin.
Roadmap
We plan to dialogue with developers in the hopes of creating an application that includes all the steps in our captioning workflow (i.e., transcription, editing, segmenting, aligning, fine-tuning) as we described in our "Needs for a Captioning Tool" document.
This application must be accessible to individuals with disabilities per WCAG 2.0 AA guidelines.
Vision
The goal is to have a partnership with eager developers who see accessible design as a priority. We also would like to provide substantial documentation that can be easily followed by people without a high level of technical expertise.
On the technical level, we hope to partner with a developer who can integrate Kaldi STT in the application because it is 1) open source and 2) amenable to customization (use of speech models).
Any blockers?
Our preferred STT tool at the moment is YouTube, which has limitations: 1) not every video uploaded will receive an auto-captions track. 2) foreign language videos may have poorer WER on YouTube compared to English videos. 3) Our use of scripts is not practical for people in our field who lack command of CLI tools. In addition, our scripts have been used primarily on the macOS and would require installation of UNIX commands on a PC (we recommend installing Git Bash).
Components
We have shared our scripts at our GitHub site and we welcome any feedback. We are not developers and anticipate more efficient and economical ways of achieving the same end.
Vision
I wish that the editing step in this workflow were faster. If I had an interface like oTranscribe but with time-codings, that would make the process faster.
Github repo if available
Github repo for the project if open source
https://github.com/polizoto/segment_transcript
https://github.com/polizoto/auto_captions_dl
Demo Video
Website
www.htctu.net
List of Stack used
Python, Perl, UNIX
Contributors list
Joseph Polizzotto (HTCTU) [email protected]
Marshall Sunnes (NYU) [email protected]