rkuchumov / ccr

Real-Time Close Captions repository

Architecture

Deployment

1. Build web pages static content: cd app && ./build.bash
2. Build containers: docker-compose build
3. Run dontainers: docker-compose up

Streaming a new channel

1. Specify database url in ./admin/config.json
2. Create a new channel (with id 'test1' and title 'Test Channel'):

./admin/ccr-channels-create -i test -i 'Test Channel'

4. Install package dependencies

cd sender && npm i

5. Start streaming:

• Using CCExtractor and real workload:

./ccr-sender -C <channel_id> -e <ccextractor_path> -u <ccextractor_source> -D <ccr-reciever_url>

• Using ccextractor mock that streams a single file

./ccr-sender -C test1 -e tests/ccextractor -u dummy_url:port  -D tcp://127.0.0.1:5000

ccr's Issues

Status Updates

Right now the only way to tell for sure if something has crashed in application is to connect to one of the servers and to check status of application’s processes.

The task is to obtain the state of CCExtractor (alive/dead, has binded to port or not) and the state of ZeroMQ connection for each channel. Then you have to store it in database and display it on the monitorng web page in real-time.

DB is stored in memory in docker-compose deployment

When CCR is deployed using docker-compose, MongoDB stores its documents in RAM instead of disk. After a dozen of days it can exhaust the memory leading to unwanted problems.

Demo website not working

I tried to visit the demo website at http://realtimedemo.ccextractor.org:8080/. All I got was an blank screen with an error message in the console.

index.js:245 Uncaught (in promise) Error: Cannot read property 'find' of undefined
    at _ (index.js:245)
    at r.<anonymous> (client.js:92)
    at r.onack (socket.js:311)
    at r.onpacket (socket.js:240)
    at r.<anonymous> (index.js:21)
    at r.emit (index.js:133)
    at r.ondecoded (manager.js:345)
    at a.<anonymous> (index.js:21)
    at a.r.emit (index.js:133)
    at a.add (index.js:241)```

Front-end framework

Web site page is implemented using React and UIKit framework. Since UIKit is plain CSS frontend framework it's not ported to React. The task is to fully replace it with Blueprint or another React-native framework (justify your choice!).

Including UIkit's less files from npm package doesn't work

Right now uikit styles are copied to imports/ui/uikit/ as I couldn't include using import.

Seems like Meteor's bug: https://forums.meteor.com/t/importing-css-from-node-module-via-less-file/29276/3

GSoC'20

Hello, I am Ridhish Jain, Web Developer, UG at IIT Dhanbad.
I found this project interesting and it was also a part of GSoC'19 ideas list.
I wanted to ask if this project can still be a part of GSoC'20??
If yes, I would love to contribute to it
If not, Is there any other way I can contribute to the organization as a web developer?

Channel Statistics

ccr-sender should also collect statistics metrics, including the following:

• Number of captions lines
• Size (in bytes) of transmitted objects
• Memory and CPU usage (of ccextractor and ccr-sender)
• suggest your own

These values should be averaged over small period of time and then displayed on a plot at web site in real-time.

Analyzing CC stream for obtaining its topic and keywords

One of Google Summer Of Code projects

Suppose you have a set of captions lines from a few minutes of newscast. When a person reads these captions he can immediately tell you which topic is being discussed, e.g. “it’s about US president election” or “it’s about a species of fish that can jump out of water ”.

We want you to do a research on how to summarize a set of given captions lines to tell a topic of them. For each method you should describe whether or not it can be applied on a stream of closed captions. Focus on advantages and disadvantages, accuracy, resources requirements (cpu, gpu, memory), implementation and maintenance difficulty, available implementations ... all in the context of closed captions.

Please, note that majority of methods you find are dedicated to document summarization, but closed captions are completely different from documents: they include dialogues, advertisements, songs lyrics and are poorly formatted (like sometimes there's no punctuation). So you should consider that in your research.

Then, when the right method is selected, you should implement it and integrate with CCR

Administration page

The only way to add, update or delete a channel is through database. Which is not convenient at all.

The task is to create a web page that would allow manage channels and captions thought API server. This page must be implemented using existing stack of technologies.

Improve message passing queue

There are some problems with the current implementation:

1. It doesn’t support authentication. For example, if the firewall is not enabled, then anyone can connect to receiver's port and start sending captions, which is obviously not desirable.
2. It doesn’t support a cluster of receivers. So if we have multiple receiver servers the admin will have to manually specify their addresses or to they'll have to use an external load balancer, which is not the best solution.

What's required to be done is either to implement these features over ZeroMQ (the required libraries are most likely already implemented) or to replace it with another, more advanced, yet as fast, queue engine.

In a new deployment, websockets are not working on a first start.

When you deploy ccr with docker-compose or aws scripts, and then start it, client don't receive real-time updates via websocket. The solution is to restart web-ui service.
I'm not sure what causing it and I don't like it.

Open source license not specified

Would you consider adding a license file to this repository, in order to make clear the conditions of use / modification?

Improve server infrastructure with Kubernetes

The task is to update Kubernetes deployment scripts (which you can find somewhere in commit history) to work with the latest ccr version and to improve them by adding data volumes, private networks, high availability and autoscaling.

Include UIkit library using "import 'uikit'"

It's possible to include src/uikit.js manually, but I didn't find a way to include uikit's components. For example, UIkit.notification('MyMessage', 'danger'); doesn't work inside components.

Maybe related:
uikit/uikit#2244
uikit/uikit#2289

Add preloading spinners to cards

The cards are initially empty when the subtitles are being fetched. This jumbles the layout and takes a bit of time for the website to take its final shape. Adding a preloader inside the cards may turn out to be helpful.