The first-ever Deepdub pipeline executing purely Text-to-Lip generation (TTL).
The driver files of this pipeline are main.py and main2.py. Why are there two, you ask? Well, it's because of weird Python packaging rules which I try very hard to get my head around, but fail every time.
Basically my original design was to keep the "modules" (in our case, Real-Time-Voice-Cloning and Wav2Lip) in the src folder, in their own respective folders. In each "module", I had created a function called run_cli. Then in the main function, I would simply call the run_cli function to do the functionality. It was clean and honest work.
But when connecting the CLI to Wooey, it gave strange errors like src/blah_blah... src is not a package.
So in pure rage, I copied the contents of each module into the same directory as the main.py driving file. It's super messy, but at least it gets the job done.
So keeping that all in mind, try using main2.py where possible because it's likely that it's the most recent version.
17/11/21: We are currently putting this pipeline aside, as is, so we can work on a complete new implementation. If it does not work well, we shall return here.
16/11/21: There are quite some things we need to improve to fix errors and perhaps streamline the process.
- Clean up code, try making into packages
- Train all models on relevant (high-res) data
- Handle error cases instead of crashing the entire pipeline on errors
- Allow user to introduce extracted/translated files by their own choice
15/11/21: Pipeline proposed improvements
