Skip to content
This repository has been archived by the owner on Apr 13, 2022. It is now read-only.

[QUESTION] human captioner(subtitler) instead of automation #40

Open
yucomds opened this issue May 7, 2021 · 3 comments
Open

[QUESTION] human captioner(subtitler) instead of automation #40

yucomds opened this issue May 7, 2021 · 3 comments

Comments

@yucomds
Copy link

yucomds commented May 7, 2021

Hi, thank for your great job.

We wonder if this solution could be adapted for the scenario of human captioners (subtitler) instead of auto-generation caption. Something like described in the below chart:

live-streaming-with-automated-multi-language-subtitling-architecture

Any subtitler could write directly to DynamoDB for example through a GUI.

what do you think about it? Do you have any suggestions?

Regards,
Yuri

@eggoynes
Copy link
Contributor

Hi Yuri,

The idea you have sounds like it could work. Replacing the automated generation with a human captioner.

Some things to keep in mind. We do not get the PTS (Presentation Time Stamp) from the video stream. Since all the video files and manifests pass through Amazon CloudFront (The TS and M3u8 files).

What this means is that the subtitles will just get updated every 2 seconds if you use 2 second segments. So roll up subtitles like in broadcast would not be possible. But you will get the subtitles updating each new segment.

You can segment what a person types into sentences, or if you want a character limit. The character limit will be a feature you will have to implement in the Lambda@Edge. Or you could limit the size of the Dynamo entries you create.

@yucomds
Copy link
Author

yucomds commented May 11, 2021

Hi @eggoynes , thanks for your answer. It's more clear now.

I worder if there is a way to get the "PTS". With this value maybe we could achive a more accurete sync. Do you think there is a way?

Regards,
Yuri

@eggoynes
Copy link
Contributor

Are you wanting to have the subtitles inserted delayed?

For the PTS question you have to get PTS from the TS video segments of the video.
CloudFront passes through all the TS video segments directly to AWS MediaPackage, so there isn't really a way to inspect them to get the PTS.

Another thing to mention is you cannot view the TS files body in Lambda@Edge since Lambda@Edge has a max 1MB total response limit.
https://stackoverflow.com/questions/55064286/lambdaedge-gives-502-only-for-some-images

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants