How can I use this project? about subtitle-synchronization HOT 4 CLOSED

I'm not that familiar with Plex but it appears it has a Python API so it shouldn't be too hard as at least you can pickle the model etc.

I want to use the approach for a VLC plug-in but I'm having difficulty getting TensorFlow to work as I have an AMD card.

from subtitle-synchronization.

Do you have to run a GPU? It's a server so its headless. If it's only . for training I might be able to help I have a GTX 970. Else you can always rent a server especially if you are a student (free sutff with your email)

from subtitle-synchronization.

Well, I wanted to learn TensorFlow but yeah I can use the CPU version - there is also hipTensorFlow which uses HIP to translate the CUDA to portable C++ and supports AMD cards so I'll try and configure that too.

Failing that I'll try and do it in the cloud - as I'm in the EU I can't get personal use for GCP but AWS is cheaper anyway. Sadly I am no longer a student 😢

from subtitle-synchronization.

This is a Proof of Concept where I have coded some test with several NN architectures and training modifications, so this is not a final API to use. However, all the needed code to process the audio and subtitles is available in this repo. Also, you can see how to create a dataset, train a NN, make predictions and synchronize a subtitles file. So, little modifications are needed to use this code.

According to the training, I use Keras with Tensorflow as backed, so you don't need to learn Ternsoflow to use this code. I expect that running one training on CPU would take a couple of hours at the most.

The training data must be stored in a 'data/' subdirectory. For each avi/mkv video, there has to be a synchronized .srt file which the same filename as its video

from subtitle-synchronization.