lattice-tool.1

这是一款很好用的工具包

lattice-tool(1)                                   lattice-tool(1)



NNAAMMEE
       lattice-tool - manipulate word lattices

SSYYNNOOPPSSIISS
       llaattttiiccee--ttooooll [--hheellpp] option ...

DDEESSCCRRIIPPTTIIOONN
       llaattttiiccee--ttooooll performs operations on word lattices in ppffssgg--
       ffoorrmmaatt(5) or in HTK Standard Lattice format (SLF).  Opera-
       tions  include size reduction, pruning, null-node removal,
       weight assignment from language models, lattice word error
       computation, and decoding of the best hypotheses.

       Each  input  lattice is processed in turn, and a series of
       optional operations  is  performed  in  a  fixed  sequence
       (regardless  of  the  order in which corresponding options
       are specified).  The sequence of operations is as follows:

       1.     Read input lattice.

       2.     Score  pronunciations (if dictionary was supplied).

       3.     Split multiword word nodes.

       4.     Posterior- and density-based pruning (before reduc-
              tion).

       5.     Write word posterior lattice.

       6.     Perform word-posterior based decoding.

       7.     Write word mesh (confusion network).

       8.     Compute  word  and  transition posteriors (forward-
              backward algorithm), and N-gram  counts  if  speci-
              fied.

       9.     Compute lattice density.

       10.    Check lattice connectivity.

       11.    Compute node entropy.

       12.    Compute lattice word error.

       13.    Output reference word posteriors.

       14.    Remove null nodes.

       15.    Lattice reduction.

       16.    Posterior-  and density-based pruning (after reduc-
              tion).

       17.    Remove pause nodes.

       18.    Lattice reduction (post-pause removal).

       19.    Language model replacement or expansion.

       20.    Pause recovery or insertion.

       21.    Lattice reduction (post-LM expansion).

       22.    Multiword splitting (post-LM expansion).

       23.    Merging of same-word nodes.

       24.    Lattice algebra operations (or, concatenation).

       25.    Viterbi-decode best hypothesis and/or  generate  N-
              best lists.

       26.    Lattice-LM perplexity computation.

       27.    Writing output lattice.

       The  following  options control which of these steps actu-
       ally apply.

OOPPTTIIOONNSS
       Each filename argument can be an ASCII  file,  or  a  com-
       pressed file (name ending in .Z or .gz), or ``-'' to indi-
       cate stdin/stdout.

       --hheellpp  Print option summary.

       --vveerrssiioonn
              Print version information.

       --ddeebbuugg _l_e_v_e_l
              Set the debugging output level (0 means  no  debug-
              ging  output).   Debugging  messages  are  sent  to
              stderr.

       --iinn--llaattttiiccee _f_i_l_e
              Read input lattice from _f_i_l_e.

       --iinn--llaattttiiccee22 _f_i_l_e
              Read additional input lattice (for  binary  lattice
              operations) from _f_i_l_e.

       --iinn--llaattttiiccee--lliisstt _f_i_l_e
              Read  list  of  input  lattices from _f_i_l_e.  Lattice
              operations are applied to each filename  listed  in
              _f_i_l_e.

       --oouutt--llaattttiiccee _f_i_l_e
              Write result lattice to _f_i_l_e.

       --oouutt--llaattttiiccee--ddiirr _d_i_r
              Write  result  lattices from processing of --iinn--llaatt--
              ttiiccee--lliisstt to directory _d_i_r.

       --rreeaadd--mmeesshh
              Assume input lattices are in word  mesh  (confusion
              network) format, as described in wwllaatt--ffoorrmmaatt(5).

       --wwrriittee--iinntteerrnnaall
              Write  output lattices with internal node numbering
              instead of compact, consecutive numbering.

       --oovveerrwwrriittee
              Overwrite existing output lattice files.

       --vvooccaabb _f_i_l_e
              Initialize the vocabulary to words listed in  _f_i_l_e.
              This is useful in conjunction with

       --lliimmiitt--vvooccaabb
              Discard  LM  parameters on reading that do not per-
              tain to the words specified in the vocabulary.  The
              default  is that words used in the LM are automati-
              cally added to the vocabulary.  This option can  be
              used  to  reduce  the memory requirements for large
              LMs; to this end, --vvooccaabb  typically  specifies  the
              set  of  words used in the lattices to be processed
              (which has to be generated  beforehand,  see  ppffssgg--
              ssccrriippttss(1)).

       --vvooccaabb--aalliiaasseess _f_i_l_e
              Reads  vocabulary alias definitions from _f_i_l_e, con-
              sisting of lines of the form
                   _a_l_i_a_s _w_o_r_d
              This causes all tokens _a_l_i_a_s to be mapped to  _w_o_r_d.

       --uunnkk   Map lattice words not contained in the known vocab-
              ulary with the unknown word tag.  This is useful if
              the  rescoring  LM  contains  a probability for the
              unknown word (i.e., is an open-vocabulary LM).  The
              known  vocabulary  is given by what is specified by
              the --vvooccaabb option, as well as all words in  the  LM
              used for rescoring.

       --mmaapp--uunnkk _w_o_r_d
              Map  out-of-vocabulary  words  to _w_o_r_d, rather than
              the default <<uunnkk>> tag.

       --ttoolloowweerr
              Map all vocabulary to lowercase.

       --nnoonneevveennttss _f_i_l_e
              Read a list of words from _f_i_l_e that are  used  only
              as  context  elements, and are not predicted by the
              LM, similar to ``<s>''.   If  --kkeeeepp--ppaauussee  is  also
              specified  then pauses are not treated as nonevents
              by default.

       --mmaaxx--ttiimmee _T
              Limit processing time per lattice to _T seconds.

       Options controlling lattice operations:

       --wwrriittee--ppoosstteerriioorrss _f_i_l_e
              Compute the posteriors of lattice nodes and transi-
              tions  (using  the  forward-backward algorithm) and
              write out a word  posterior  lattice  in  wwllaatt--ffoorr--
              mmaatt(5).   This and other options based on posterior
              probabilities make most sense if the input  lattice
              contains  combined acoustic-language model weights.

       --wwrriittee--ppoosstteerriioorrss--ddiirr _d_i_r
              Similar to the above, but  posterior  lattices  are
              written  to  separate files in directory _d_i_r, named
              after the utterance IDs.

       --wwrriittee--mmeesshh _f_i_l_e
              Construct a word confusion network ("sausage") from
              the  lattice  and  write  it to _f_i_l_e.  If reference
              words are available for the utterance (specified by
              --rreeff--ffiillee  or  --rreeff--lliisstt)  their  alignment will be
              recorded in the sausage.

       --wwrriittee--mmeesshh--ddiirr _d_i_r
              Similar, but write sausages to files in  _d_i_r  named
              after the utterance IDs.

       --iinniitt--mmeesshh _f_i_l_e
              Initialize the word confusion network by reading an
              existing  sausage  from  _f_i_l_e.   This   effectively
              aligns  the lattice being processed to the existing
              sausage.

       --aaccoouussttiicc--mmeesshh
              Preserve word-level  acoustic  information  (times,
              scores, and pronunciations) in sausages, encoded as
              described in wwllaatt--ffoorrmmaatt(5).

       --ppoosstteerriioorr--pprruunnee _P
              Prune lattice nodes with  posteriors  less  than  _P
              times the highest posterior path.

       --ddeennssiittyy--pprruunnee _D
              Prune  lattices such that the lattice density (non-
              null words per second) does not exceed _D.

       --nnooddeess--pprruunnee _N
              Prune lattices such that the total number  of  non-
              null, non-pause nodes does not exceed _N.

       --ffaasstt--pprruunnee
              Choose  a  faster  pruning  algorithm that does not
              recompute posteriors after each iteration.

       --wwrriittee--nnggrraammss _f_i_l_e
              Compute posterior expected N-gram  counts  in  lat-
              tices  and output them to _f_i_l_e.  The maximal N-gram
              length is given by the --oorrddeerr option  (see  below).
              The  counts from all lattices processed are accumu-
              lated and output at the end.

       --wwrriittee--nnggrraamm--iinnddeexx _f_i_l_e
              Output an index file of all  N-gram  occurences  in
              the   lattices  processed,  including  their  start
              times, durations, and posterior probabilities.  The
              maximal N-gram length is given by the --oorrddeerr option
              (see below).

       --mmiinn--ccoouunntt _C
              Prune N-grams with count less than  _C  from  output
              with  --wwrriittee--nnggrraammss and --wwrriittee--nnggrraamm--iinnddeexx.  In the
              former case, the threshold applies to the aggregate
              occurrence  counts; in the latter case, the thresh-
              old applies to  the  posterior  probability  of  an
              individual occurence.

       --mmaaxx--nnggrraamm--ppaauussee _T
              Index  only  N-grams  that  contain internal pauses
              (between words) not exceeding _T  seconds  (assuming
              time stamps are recorded in the input lattice).

       --ppoosstteerriioorr--ssccaallee _S
              Scale  the  transition weights by dividing by _S for
              the purpose of posterior  probability  computation.
              If  the  input weights represent combined acoustic-
              language model scores then this should be  approxi-
              mately  the language model weight of the recognizer
              in order to avoid  overly  peaked  posteriors  (the
              default value is 8).

       --wwrriittee--vvooccaabb _f_i_l_e
              Output  the  list  of  all  words found in the lat-
              tice(s) to _f_i_l_e.

       --rreedduuccee
              Reduce lattice size by a single forward node  merg-
              ing pass.

       --rreedduuccee--iitteerraattee _I
              Reduce  lattice  size  by  up to _I forward-backward
              node merging passes.

       --oovveerrllaapp--rraattiioo _R
              Perform approximate lattice  reduction  by  merging
              nodes  that  share  more than a fraction _R of their
              incoming or outgoing  nodes.   The  default  is  0,
              i.e., only exact lattice reduction is performed.

       --oovveerrllaapp--bbaassee _B
              If  _B  is 0 (the default), then the overlap ratio _R
              is taken relative to the smaller set of transitions
              being  compared.   If  the value is 1, the ratio is