CTC Alignment implemented with CUDA

What does this module do?

To understand the function of this module, you must acknowledge the CTC first. Have a look at the references if you're new to the term CTC.

I recently ran into a problem that required to cast sequences with CTC alignment. The process of the algorithm is straight-forward and simple:

1. Squeeze consecutive repetitive symbols to one: 1 1 1 2 2 3 3 3 3 -> 1 2 3;
2. Remove all blank symbols (commonly we use index 0 to identify that): 1 0 2 0 0 3 -> 1 2 3.

e.g. 1 0 1 2 2 0 0 3 0 4 -> 1 1 2 3 4

However, things become complicated when there're many many long sequences. Nested loop would consume intolerable computing time. Therefore I decided to implement the alignment algo. in CUDA.

Install

python setup.py install
# or install locally
# python setup.py develop

Usage

Please refer to the source code.

Reference

1. A brief intro to CTC. https://distill.pub/2017/ctc/