Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks

This paper introduces Connectionist Temporal Classification (CTC), a method for training recurrent neural networks to label sequential data without needing pre-segmented training data. CTC allows the network to learn alignments between the input sequence and the labels, making it suitable for tasks like speech recognition where precise alignment is unknown. The approach outputs a probability distribution over all possible label sequences for a given input, and it is trained using a maximum likelihood approach. This eliminates the need for pre-processing or post-processing steps for alignment.