The seminal paper from Alex Graves and Jürgen A. Schmidhuber that introduced CTC, Connectionist temporal classification labelling unsegmented sequence data with recurrent neural networks, is moderately readable.
Sequence Modeling with CTC from distill.pub (written in 2017) is the best resource for understanding Connectionist Temporal Classification intuitively.
There is also a decent blog post from Harald Scheidl that also explains CTC: An Intuitive Explanation of Connectionist Temporal Classification. More usefully from him, there are CTC Decoding Algorithms implemented in the githubharald/CTCDecoder repo. There is also the CTC Word Beam Search Decoding Algorithm implementation repo.
The PyTorch CTCLoss is a solid reference implementation.