“Sequence Modeling With CTC: A Visual Guide to Connectionist Temporal Classification, an Algorithm Used to Train Deep Neural Networks in Speech Recognition, Handwriting Recognition and Other Sequence Problems”, 2017-11-27 (; similar):
Thorough and heavily-illustrated explanation of Connectionist temporal classification (CTC), a way to grade the quality of any sequence-to-sequence problem, such as text-to-speech or speech transcription.
Because there are many possible sequences which mean similar things but may be completely unaligned, such problems cannot be tackled by the usual classification loss; CTC turns out to be an elegant general solution, efficiently computable by dynamic programming, and surfacing in many apparently unrelated problems.
In particular, CTC is essential to training powerful neural networks and is one reason why voice-related tasks have seen such large performance gains in the past decades.