athena.tools.lm_scorer

language model scorer used in speech recognition model decoding

Module Contents

Classes

NGramScorer	KenLM language model
RNNScorer	RNN language model

class athena.tools.lm_scorer.NGramScorer(lm_path, sos, eos, num_syms, lm_weight=0.1)

Bases: object

KenLM language model

reset(self)

Call this function to reset the lm to predict on a new sequence

score(self, candidate_holder, new_scores)

Call this function to compute the NGram score of the next prediction based on historical predictions, the scoring function shares a common interface :param candidate_holder:

Returns:the NGram weighted score cand_states: Return type:score

get_score(self, cand_parents, cand_syms, lang_model)

the saved lm model will be called here :param cand_parents: last selected top candidates :param cand_syms: last selected top char index :param lang_model: the language model

Returns:the lm scores Return type:scores

class athena.tools.lm_scorer.RNNScorer(lm_model, lm_weight=0.1)

Bases: object

RNN language model

reset(self)

Call this function to reset the lm to predict on a new sequence

score(self, candidate_holder, new_scores)

Call this function to compute the score of the next prediction based on historical predictions, the scoring function shares a common interface :param candidate_holder:

Returns:the weighted score cand_states: Return type:score