kaldi-lattice-word-index

This tool builds a word index from character lattices.

Build a word index from character lattices.

Words are any sequence of characters in between any of the separator symbols (e.g.: whitespace, punctuation marks, etc).

The program will output the n-best character segmentations of words, with their scores. More precisely:

Let's define a binary variable R that denotes whether the character transcript (y) of the sample (x) contains the word formed by the sequence of characters c, where each character is segmented according to the sequence s.

Then, the program computes:

• If --only-best-segmentation=false (the default) then:

• If --only-best-segmentation=true then:

In any case, the score for a character sequence (c) is a lower bound to P(R = 1 | x, c), but it is usually quite close.

Usage

Usage: kaldi-lattice-word-index [options] separator-symbols lat-rspecifier
 e.g.: kaldi-lattice-word-index "1 2" ark:lats.ark
 e.g.: kaldi-lattice-word-index --nbest=10000 "1 2" ark:lats.ark

Options

• --acoustic-scale : Scaling factor for acoustic likelihoods in the lattices. (float, default = 1)
• --beam : Pruning beam (applied after acoustic scaling and adding the insertion penalty). (float, default = inf)
• --graph-scale : Scaling factor for graph probabilities in the lattices. (float, default = 1)
• --insertion-penalty : Add this penalty to the lattice arcs with non-epsilon output label (typically, equivalent to word insertion penalty). (float, default = 0)
• --max-mem : Maximum approximate memory usage in determinization (real usage might be many times this). (int, default = 536870912)
• --nbest : Extract this number of n-best hypothesis. (int, default = 100)
• --only-best-segmentation : If true, output the best character segmentation for each word. (bool, default = false)
• --symbols-table : Use this symbols table to map from labels to characters. (string, default = "")