Understanding peephole connections

The point behind peephole connections is the need to capture the information of time lags. In other words, we wish to include the information conveyed by time intervals between sub-patterns of sequences in our modeling efforts. This is relevant not only for certain language processing tasks (such as speech recognition), but also for numerous other tasks ranging from machine motor control to maintaining elaborate rhythms in computer-generated music. Previous approaches to tasks such as speech recognition employed the use of Hidden Markov Models (HMMs). These are essentially statistical models that estimate the probability of a set of observations based on the sequence of hidden state transitions. In the case of speech processing, observations are defined as segments of digital signals corresponding to speech, while Markov hidden states are the...