nbest-optimize.1

这是一款很好用的工具包

nbest-optimize(1)                               nbest-optimize(1)



NNAAMMEE
       nbest-optimize  -  optimize  score  combination for N-best
       word error minimization

SSYYNNOOPPSSIISS
       nnbbeesstt--ooppttiimmiizzee [--hheellpp] option ...  [ _s_c_o_r_e_d_i_r ...  ]

DDEESSCCRRIIPPTTIIOONN
       nnbbeesstt--ooppttiimmiizzee reads a set  of  N-best  lists,  additional
       score  files,  and corresponding reference transcripts and
       optimizes the score combination weights so as to  minimize
       the  word  error  of a classifier that performs word-level
       posterior probability maximization.  The optimized weights
       are  meant  to be used with nnbbeesstt--llaattttiiccee(1) and the --uussee--
       mmeesshh  option,  or  the  nnbbeesstt--rroovveerr  script  (see   nnbbeesstt--
       ssccrriippttss(1)).  nnbbeesstt--ooppttiimmiizzee determines both the best rel-
       ative weighting of knowledge source scores and the optimal
       --ppoosstteerriioorr--ssccaallee parameter that controls the peakedness of
       the posterior distribution.

       The optimization is performed by  gradient  descent  on  a
       smoothed  (sigmoidal)  approximation  of the true 0/1 word
       error function (Katagiri et  al.  1990).   Therefore,  the
       result  can  only be expected to be a _l_o_c_a_l minimum of the
       error surface.  (A more global search can be attempted  by
       specifying different starting points.)  Another approxima-
       tion is that the error function  is  computed  assuming  a
       fixed  multiple alignment of all N-best hypotheses and the
       reference string, which tends to slightly overestimate the
       true  pairwise error between any single hypothesis and the
       reference.

       An  alternative  search  strategy  uses  a   simplex-based
       "Amoeba"  search on the (non-smoothed) word error function
       (Press et al. 1988).  The  search  is  restarted  multiple
       times to avoid local minima.

       Alternatively,  nnbbeesstt--ooppttiimmiizzee  can  also optimize weights
       for a standard, 1-best hypothesis rescoring  that  selects
       entire  (sentence)  hypotheses  (--11bbeesstt  option).  In this
       mode sentence-level error counts may be read from external
       files,  or computed on the fly from the reference strings.
       The weights obtained are meant to be used for N-best  list
       rescoring with rreessccoorree--rreewweeiigghhtt (see nnbbeesstt--ssccrriippttss(1)).

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
              Controls  the  amount  of  output  (the  higher the
              _l_e_v_e_l, the more).  At level 1, error statistics  at
              each  iteration  are  printed.   At  level  2, word
              alignments are printed.  At  level  3,  full  score
              matrix  is  printed.  At level 4, detailed informa-
              tion about word hypothesis ranking is  printed  for
              each training iteration and sample.

       --nnbbeesstt--ffiilleess _f_i_l_e_-_l_i_s_t
              Specifies  the  set  of  N-best  files as a list of
              filenames.   Three  sets  of  standard  scores  are
              extracted from the N-best files: the acoustic model
              score, the language model score, and the number  of
              words  (for  insertion  penalty  computation).  See
              nnbbeesstt--ffoorrmmaatt(5) for details.

       --rreeffss _r_e_f_e_r_e_n_c_e_s
              Specifies the reference transcripts.  Each line  in
              _r_e_f_e_r_e_n_c_e_s  must  contain the sentence ID (the last
              component in the N-best filename  path,  minus  any
              suffixes) followed by zero or more reference words.

       --iinnsseerrttiioonn--wweeiigghhtt _W
              Weight insertion errors by a factor _W.  This may be
              useful to optimize for keyword spotting tasks where
              insertions have a cost different from deletion  and
              substitution errors.

       --wwoorrdd--wweeiigghhttss _f_i_l_e
              Read  a table of words and weights from _f_i_l_e.  Each
              word error is weighted according to  the  word-spe-
              cific weight.  The default weight is 1, and used if
              a word has no specified weight.   Also,  when  this
              option  is used, substitution errors are counted as
              the sum of a deletion and an  insertion  error,  as
              opposed  to  counting  as 1 error as in traditional
              word error computation.

       --11bbeesstt Select  optimization  for  standard  sentence-level
              hypothesis selection.

       --11bbeesstt--ffiirrsstt
              Optimized  first  using --11bbeesstt mode, then switch to
              full optimization.  This is  an  effective  way  to
              quickly  bring  the  score  weights near an optimal
              point, and then fine-tune  them  jointly  with  the
              posterior scale parameter.

       --eerrrroorrss _d_i_r
              In  1-best mode, optimize for error counts that are
              stored in separate files in directory _d_i_r.  Each N-
              best list must have a matching error counts file of
              the same basename in _d_i_r.   Each  file  contains  7
              columns of numbers in the format
                   wcr wer nsub ndel nins nerr nw
              Only  the  last  two  columns (number of errors and
              words, respectively) are used.
              If this option is omitted, errors will be  computed
              from  the N-best hypotheses and the reference tran-
              scripts.

       --mmaaxx--nnbbeesstt _n
              Limits the number of hypotheses read from  each  N-
              best list to the first _n.

       --rreessccoorree--llmmww _l_m_w
              Sets  the  language  model weight used in combining
              the language model log probabilities with  acoustic
              log probabilities.  This is used to compute initial
              aggregate hypotheses scores.

       --rreessccoorree--wwttww _w_t_w
              Sets the word transition weight used to weight  the
              number of words relative to the acoustic log proba-
              bilities.  This is used to compute  initial  aggre-
              gate hypotheses scores.

       --ppoosstteerriioorr--ssccaallee _s_c_a_l_e
              Initial  value  for  scaling  log  posteriors.  The
              total weighted log score is divided by  _s_c_a_l_e  when
              computing normalized posterior probabilities.  This
              controls the peakedness of the posterior  distribu-
              tion.  The default value is whatever was chosen for
              --rreessccoorree--llmmww, so that  language  model  scores  are
              scaled  to  have weight 1, and acoustic scores have
              weight 1/_l_m_w.

       --ccoommbbiinnee--lliinneeaarr
              Compute aggregate  scores  by  linear  combination,
              rather   than  log-linear  combination.   (This  is
              appropriate if the input scores represent  log-pos-
              terior probabilities.)

       --nnoonn--nneeggaattiivvee
              Constrain search to non-negative weight values.

       --vvooccaabb _f_i_l_e
              Read  the  N-best  list vocabulary from _f_i_l_e.  This
              option is mostly redundant since words found in the
              N-best  input  are  implicitly added to the vocabu-
              lary.

       --ttoolloowweerr
              Map vocabulary to lowercase, eliminating case  dis-
              tinctions.

       --mmuullttiiwwoorrddss
              Split  multiwords  (words joined by '_') into their
              components when reading N-best lists.