// doc/glossary.dox

// Copyright 2015 Johns Hopkins University (author: Daniel Povey)

// See ../../COPYING for clarification regarding multiple authors

//

// Licensed under the Apache License, Version 2.0 (the "License");

// you may not use this file except in compliance with the License.

// You may obtain a copy of the License at

// http://www.apache.org/licenses/LICENSE-2.0

// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY

// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED

// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,

// MERCHANTABLITY OR NON-INFRINGEMENT.

// See the Apache 2 License for the specific language governing permissions and

// limitations under the License.

/**

\page glossary Glossary of terms

This page contains a glossary of terms that Kaldi users might want to know

about. The current content here consists just of a few examples; more content

will be added shortly. The easiest way to search in this page is to use the

search function of your browser. For convenience the definition of each

term's section is preceded and followed by a colon, so for

instance, typing ctrl-f ":lattice:" would take you to the section for "lattice".

<div style="text-indent: -1.5em; padding-left: 1.5em;">

<b>:acoustic scale:</b> The acoustic scale used in decoding, written as --acoustic-scale

in C++ programs and --acwt in programs. This is a scale on the acoustic log-probabilities,

and is a universally used kludge in HMM-GMM and HMM-DNN systems to account for the correlation

between frames. It's usually set to 0.1, meaning the acoustic log-probs get a much lower weight

than the language model log-probs. In scoring scripts you'll often see a range of language

model weights being searched over (like the range 7 to 15). These can be interpreted as

the inverse of an acoustic scale; it's the ratio between the two that matters for Viterbi

decoding.

<b>:alignment:</b> A representation of the sequence of HMM states taken by the

Viterbi (best-path) alignment of an utterance. In Kaldi an alignment is

synonymous with a sequence of <b>transition-ids</b>. Most of the time an

alignment is derived from aligning the reference transcript of an utterance,

in which case it is called a <b>forced alignment</b>. <b>lattices</b> also

contain alignment information as sequences of transition-ids for each word

sequence in the lattice. The program \ref bin/show-alignments.cc "show-alignments" shows

alignments in a human-readable format.

<b>:cost:</b> Any quantity which is used as a 'cost' in a weighted FST

algorithm (e.g. acoustic cost, language model cost; see \ref lattices

for more details). Costs are, generally speaking, interpretable

as a negative log of a likelihood or probability, but there may

be scaling factors involved.

<b>:forced alignment:</b> see <b>alignment</b>.

<b>:lattice:</b> A representation of alternative likely transcriptions of an utterance, together

with associated alignment and cost information. See \ref lattices.

<b>:likelihood:</b> A mathematical concept meaning the value of a function representing

the distribution of a continuous value. These can be more than one. Often represented

in log space (as log-likelihood) because likelihood values of multi-dimensional

features can often be too small or large to fit in standard floating-point precision.

With standard cross-entropy trained neural net systems we obtain "pseudo-likelihoods"

by dividing log-probabilities by the priors of context-dependent states.

<b>:posterior:</b> "Posterior" is shorthand for "posterior probability" which is a very

general mathematical concept, generally meaning "the probability of some random

variable after seeing the relevant data". In general, posteriors will sum to one.

In Kaldi terminology, if you encounter the term "posterior", abbreviated to "post",

without further expanation it generally means the per-frame posterior probability of

transition-ids. However these posteriors may be very peaky (i.e. mostly ones and zeros)

depending how you obtained them, e.g. from a lattice or from an alignment.

Alignments and lattices can be converted to posteriors over transition-ids (see \ref lattice-to-post.cc),

or over lattice arcs (see \ref ali-to-post.cc and \ref lattice-arc-post.cc).

Posteriors over transition-ids can be converted to posteriors over pdf-ids or over phones;

see the tools \ref ali-to-post.cc, \ref post-to-pdf-post.cc and \ref post-to-phone-post.cc.

<b>:pdf-id:</b> The zero-based integer index of a clustered context-dependent HMM state; see

\ref transition_model_identifiers for more information.

<b>:transition-id:</b> a one-based index that encodes the pdf-id (i.e. the clustered context-dependent HMM state),

the phone identity, and information about whether we took the self-loop or forward transition in the HMM.

Appears in lattices, decoding graphs and alignments. See \ref transition_model.

<b>:transition model:</b> The TransitionModel object encodes the transition probabilities

of the HMMs, and also various other important integer mappings; see \ref transition_model.

This object is generally written at the start of model files. The program

\ref bin/show-transitions.cc "show-transitions" shows these.

<b>:G.fst:</b> The grammar FST <code>G.fst</code> which lives in the

<code>data/lang/</code> directory in the scripts (see \ref data_prep_lang) represents

the language model in a Finite State Transducer format (see www.openfst.org).

For the most part it is an acceptor, meaning the input and output symbols on the

arcs are the same, but for statistical language models with backoff, the backoff

arcs have the "disambiguation symbol" <code>#0</code> on the input side only.

For many purposes you'll want to get rid of the disambiguation symbols

using the command <code>fstproject --project_output=true</code>. The disambiguation symbols

are needed during graph compilation to make the FST determinizable, but for things

like language-model rescoring you don't want them.

</div>

*/