Loads a data file but stores input frames in a dictionary keyed by utterance Each input dictionary entry is a 2-D matrix of length eq

Forward prop entire utterance Call CTC cost function Compute gradient data is a 2-D matrix where each column is a si

Initialize parameters using 6/sqrt(fanin+fanout)

Initialize parameters using 6/sqrt(fanin+fanout)

CTC loss function. params - n x m matrix of n-D probability distributions over m frames. seq - sequence of phone id's for given example.

Initialize parameters using 6/sqrt(fanin+fanout)

Edit distance between two sequences reference (ref) and hypothesis (hyp). Returns edit distance, number of insertions, deletions and substitu

Initialize parameters using 6/sqrt(fanin+fanout)

Loads a data file but stores input frames in a dictionary keyed by utterance Each input dictionary entry is a 2-D matrix of length eq

Runs stochastic gradient descent with nesterov acceleration. Model is objective.

Forward prop entire utterance Call CTC cost function Compute gradient data is a 2-D matrix where each column is a si

Loads file with label to integer mapping. Use unique_tokens to create file.

Remove utterance ids and lowercase

Convert raw text into character language model format Split characters up, add start and end symbols, replace spaces Avoid splitting up c

Saves only the network parameters to the given fd.

Saves only the network parameters to the given fd.

Converts a stack object into a single parameter vector x is a stack returns a single numpy array XXX or does this ret