class BeamSearchDecoder

This class implements a Viterbi decoder with beam search capabilities.

Public Members

Decodes using the input data vectors in 'input_data'

'n_frames_' is the number of vectors of input data

'vec_size' is the number of elements in each vector.

features or emission probabilities and 'vec_size' must reflect this

After this function returns, 'num_result_words' contains the number of words

recognised and 'result_words' contains the vocabulary indices of the recognised

Documentation

This class implements a Viterbi decoder with beam search capabilities. A Lexicon and LanguageModel are required at creation time (the LanguageModel is optional). By default, no pruning occurs. Two levels of pruning can be configured - word interior hypothesis pruning and word end hypothesis pruning. The application of language model probabilities can be delayed or performed normally.

Decodes using the input data vectors in 'input_data'

Decodes using the input data vectors in 'input_data'

'n_frames_' is the number of vectors of input data

'n_frames_' is the number of vectors of input data

'vec_size' is the number of elements in each vector.

'vec_size' is the number of elements in each vector. The input data can be either

features or emission probabilities and 'vec_size' must reflect this

features or emission probabilities and 'vec_size' must reflect this

After this function returns, 'num_result_words' contains the number of words

After this function returns, 'num_result_words' contains the number of words

recognised and 'result_words' contains the vocabulary indices of the recognised

recognised and 'result_words' contains the vocabulary indices of the recognised

This class has no child classes.

Author:

Darren Moore (moore@idiap.ch)

Alphabetic index HTML hierarchy of classes or Java