Overview:
Media Machine aims at detecting mismatch between original and dubbed movie file. The mismatch can be of any type between the original and dubbed files like Faster, Slower, Left-Shift, Right-Shift etc.
The Challenge:
The movie industries provide dubbed version of their movie for almost all region in the world. But due to various region specific challenges, (Sensor board rules for content, language etc.) making dubbed version takes time and consume resources. So, the movie making companies usually hire third party from the specific region to make the dubbed version of the movie.
Our media machine project aims to address the following challenges:
- Detect mismatch for Faster or Slower case which is generate due to the language dependent sentence length difference.
- Detect mismatch for Left-Shift or Right-Shift which is generate due to additional scene added in compliance with the the sensor board rule.
The objective was to make a software which will be capable of detecting the above mentioned mismatches. If those mismatches can be found, then they can be reported back for correction
At present, the industries need to check the dubbed version by hiring a person who will sit in front of the TV and watch the whole dubbed movie with the original one side by side. But after this lengthy process of observation, the mismatch report by human can still have issue which we have noticed in some movies.
The Solution:
Pridesys Global LLC. provide the following options to overcome above challenges:
Mismatch Detection AI:
We’ve used deep learning framework to make the core of mismatch prediction module. The AI will predict the mismatch with more than 93% accuracy. We’ve trained our Machine learning model with more than Half a millions of data extracting from 25 movies. We’ve used GPU based learning process to make our learning process faster.
Mismatch Reporting Module:
We’ve created a reporting module which will provide users with the summary of the whole movie. The details of mismatch prediction data can also be shown if the user asks for.
Movie Process from AWS S3 or Local:
Due to large size of movies, uploading can be bandwidth and time-consuming, So we’ve added the facility of processing data directly from AWS S3.
The Result:
Using Media machine, any movie making company will be able to:
Detect mismatch portion of movie very quickly:
Using Media machine, the process of detecting fault in the dubbed movie is quicker that anyone could imagine. Whole move can be processed in just few minutes. All the user need to do is just jumps to the mismatch location of the movie to see what’s happening there.
Report mismatch issue to Third part easily:
User don’t have to report the mismatch manually. They just need to export the report after the process and it is then ready to be sent to the dubbed making Third party.
Technology Used:
We’ve used the following technologies for the Media machine project:
MXNet: Deep learning model creation for mismatch detection AI.
Python: Programming language for the application.
Flask: Light python-based web framework for user interface.
AWS: We’ve use AWS as the platform for the application.