cat

speech signal process tools

REFCOF(1-ESPS)                              ENTROPIC RESEARCH LAB



NNNNAAAAMMMMEEEE
     refcof - computes LPC reflection coefficients via various spectrum analysis methods

     xrefcof- run _r_e_f_c_o_f with X Windows interactions and displays

SSSSYYYYNNNNOOOOPPPPSSSSIIIISSSS
     rrrreeeeffffccccooooffff [ ----PPPP _p_a_r_a_m ] [ ----pppp _r_a_n_g_e ] [ ----rrrr _r_a_n_g_e ] [ ----llll _f_r_a_m_e__l_e_n
     ]  [ ----SSSS _s_t_e_p ] [ ----wwww _w_i_n_d_o_w__t_y_p_e ] [ ----mmmm _m_e_t_h_o_d ] [ ----oooo _o_r_d_e_r ]
     [ ----eeee _p_r_e_e_m_p_h_a_s_i_s ] [ ----cccc _c_o_n_v__t_e_s_t ] [ ----iiii _m_a_x__i_t_e_r ]  [   ---- ssss
     _s_i_n_c__n  ]  [   ----dddd ] [ ----ZZZZ ] [ ----zzzz ] [ ----xxxx _d_e_b_u_g__l_e_v_e_l ] _f_i_l_e._s_d
     _f_i_l_e._r_c

     xxxxrrrreeeeffffccccooooffff [ ----{{{{rrrrpppp}}}} _r_a_n_g_e ] _s_d__f_i_l_e


DDDDEEEESSSSCCCCRRRRIIIIPPPPTTTTIIIIOOOONNNN
     _r_e_f_c_o_f takes an ESPS sampled data file,  _f_i_l_e._s_d,  and  pro-
     duces  an  ESPS FEA_ANA analysis file _f_i_l_e._r_c containing the
     reflection coefficients corresponding to one or more  fixed-
     length sampled-data frames.

     All input frames have the same length  _f_r_a_m_e__l_e_n  (see   ---- llll
     option).  The initial point of the first frame is determined
     by the ----pppp option or by _s_t_a_r_t in the parameter file.  Initial
     points  of  any  subsequent frames follow at equal intervals
     _s_t_e_p (see ----SSSS option).  Thus the 3 cases  _s_t_e_p  <  _f_r_a_m_e__l_e_n,
     _s_t_e_p  = _f_r_a_m_e__l_e_n, and _s_t_e_p > _f_r_a_m_e__l_e_n, correspond to over-
     lapping frames, exactly abutted frames, and frames separated
     by gaps.

     The number of frames is the minimum sufficient  to  cover  a
     specified range of _n_a_n points (see ----pppp and ----ZZZZ options), given
     _f_r_a_m_e__l_e_n and _s_t_e_p.  The last frame in each file may overrun
     the  range,  in which case a warning is printed.  If a frame
     overruns the end of a file, it is normally filled  out  with
     zeros (but see ----ZZZZ).

     The reflection coefficients, along with the computed  values
     for  _r_a_w__p_o_w_e_r  and  _l_p_c__p_o_w_e_r,  are  then stored in FEA_ANA
     records.  No pitch pulse information is written to the file.

     If _f_i_l_e._s_d is "-" then the input is read from  the  standard
     input  and  if _f_i_l_e._r_c is "-" then the output is directed to
     the standard output.

     The following spectrum analysis methods are available:
9         Autocorrelation Method (AUTOC) - see _g_e_t__a_u_t_o (3-ESPS)
9         Covariance Method (COV) - see _c_o_v_a_r (3-ESPS)
9         Burg Method (BURG) - see _g_e_t__b_u_r_g (3-ESPS)
9         Modified Burg Method (MBURG) - see _g_e_t__b_u_r_g (3-ESPS)



        Copyright Entropic Research Laboratory, Inc. 1G1        1






REFCOF(1-ESPS)                              ENTROPIC RESEARCH LAB



         Fast Modified Burg Method (FBURG) - _g_e_t__f_b_u_r_g (3-ESPS)
9         Structured Covariance Method (STRCOV and STRCOV1) - see _b_e_s_t_a_u_t_o (3-ESPS),
         _s_t_r_u_c_t__c_o_v (3-ESPS), and _g_e_n_b_u_r_g (3-ESPS)
9         Vector Burg Method (VBURG) (fast approximation to structured
         covariance)  - see _g_e_t__v_b_u_r_g (3-ESPS)
9
     By default, the program  uses  the  autocorrelation  method,
     which  applies  the  standard  method of computing the auto-
     correlation function and the Levinson algorithm for  comput-
     ing  the  reflection  coefficients  from the autocorrelation
     values.  Note that the standard autocorrelation  method  may
     yield poor results when using a small number (i.e. < 100) of
     data samples.  The other methods can be specified  by  using
     the  ---- mmmm option.  The most accurate usually is STRCOV, espe-
     cially for small frame lengths.

     The methods AUTOC, STRCOV, and STRCOV1 operate by estimating
     the   autocorrelation  function  and  then  transforming  to
     reflection coefficients.  In these cases,  the  program  can
     also  optionally  multiply the autocorrelation function by a
     sinc function (----ssss option) prior to computing the  reflection
     coefficients.  This  has the effect of reducing the spectral
     resolution if the spectrum of these coefficients is plotted.

     Of the two structured  covariance  methods  [2],  STRCOV  is
     consderably  faster  and  better  behaved  than STRCOV1.  We
     include STRCOV1 as  it  may  be  useful  in  certain  cases.
     STRCOV  uses a fast, single channel algorithm _s_t_r_u_c_t__c_o_v (3-
     ESPS) developed by  John  Burg  and  programmed  by  Bernard
     Fraenkel.   STRCOV2  uses  an older (but more general) algo-
     rithm _g_e_n_b_u_r_g (3-ESPS) that was programmed by Daniel Wenger.
     Note that the ----cccc and ----iiii options are relevant for controlling
     the convergence of STRCOV.   The  VBURG  method  is  a  fast
     approximation  to  structured  covariance that was developed
     and programmed by John Burg and Shankar Narayan [3].

     If spectral representations other  than  reflection  coeffi-
     cients  are  desired, use _t_r_a_n_s_p_e_c (1-ESPS) or _s_p_e_c_t_r_a_n_s (1-
     ESPS) on the output of _r_e_f_c_o_f. If you want the actual  spec-
     trum, use _m_e__s_p_e_c (1-ESPS) on the output of _r_e_f_c_o_f.

     _x_r_e_f_c_o_f is a script that runs _r_e_f_c_o_f on a  single  frame  of
     data  that is specified by the range option (----rrrr or ----pppp) or by
     means of ESPS Common.  A pop-up window is used to prompt the
     user   for   _w_i_n_d_o_w__t_y_p_e,   _m_e_t_h_o_d,  _o_r_d_e_r,  _c_o_n_v__t_e_s_t,  and
     _m_a_x__i_t_e_r.  The results of the _a_n_a_l_y_s_i_s are displayed in  two
     pop-up  windows - one containing the reflection coefficents,
     and one containing a maximum-entropy power spectrum computed
     from  these reflection coeffiecients.  _x_r_e_f_c_o_f makes used of
     _e_x_p_r_o_m_p_t (1-ESPS), _m_e__s_p_e_c (1-ESPS), _p_l_o_t_s_p_e_c (1-ESPS),  and
     _x_t_e_x_t (1-ESPS).



        Copyright Entropic Research Laboratory, Inc. 2G2        2


9


REFCOF(1-ESPS)                              ENTROPIC RESEARCH LAB



     The parameter prompting for _x_r_e_f_c_o_f is performed by means of
     the   parameter  file  named  PWrefcof,  which  is  normally
     obtained from $ESPS_BASE/lib/params.  However, if you have a
     file by this name in the current directory (or if you define
     the environment variable ESPS_PARAMS_PATH  and  put  one  on
     that path), it will be used instead.

OOOOPPPPTTTTIIIIOOOONNNNSSSS
     The following options are supported (only ----rrrr or  ----pppp  can  be
     given for _x_r_e_f_c_o_f):

     ----PPPP _p_a_r_a_m
          uses the parameter file _p_a_r_a_m rather than the  default,
          which is _p_a_r_a_m_s.

     ----pppp _f_i_r_s_t::::_l_a_s_t

     ----pppp _f_i_r_s_t-_l_a_s_t

     ----pppp _f_i_r_s_t::::++++_i_n_c_r [1:+999]
          In the first two forms, a  pair  of  unsigned  integers
          specifies  the  range  of  sampled data to analyze.  If
          _l_a_s_t = _f_i_r_s_t + _i_n_c_r, the  third  form  (with  the  plus
          sign)  specifies the same range as the first two forms.
          If _f_i_r_s_t is omitted, the default value of  1  is  used.
          If _l_a_s_t is omitted, then a default frame length of 1000
          results.  If the specified range contains points not in
          the file, the last frame is truncated to fit the actual
          data.  Both forms of the option override the values  of
          _s_t_a_r_t  and  _n_a_n  in  the  parameter file or ESPS Common
          file.  If the ----pppp option  is  not  used,  the  range  is
          determined  from  the  ESPS Parameter or Common file if
          the appropriate parameters are present.  Note that  the
          default frame length of 1000 also results if _n_a_n is not
          in the parameter or Common file and if no ----pppp is used.

     ----rrrr _r_a_n_g_e
          ----rrrr is a synonym for ----pppp.

     ----llll _f_r_a_m_e__l_e_n [0]
          Specifies the length of each frame.  If the  option  is
          omitted, the parameter file is consulted.  A value of 0
          (from either the option or the  parameter  file)  indi-
          cates  that  a  single  frame of length _n_a_n (see ----pppp) is
          processed; this is  also  the  default  value  in  case
          _f_r_a_m_e__l_e_n is not specified either with the ----llll option or
          in the parameter file.

     ----SSSS _s_t_e_p [_f_r_a_m_e__l_e_n]
          Initial points of consecutive  frames  differ  by  this
          number  of  samples.   If  the  option  is omitted, the
          parameter file is consulted, and if no value  is  found



        Copyright Entropic Research Laboratory, Inc. 3G3        3






REFCOF(1-ESPS)                              ENTROPIC RESEARCH LAB



          there,  a default equal to _f_r_a_m_e__l_e_n is used (resulting
          in exactly abutted frames).  (The same default  applies
          if _s_t_e_p is given a value of 0).

     ----wwww _w_i_n_d_o_w__t_y_p_e[RECT]
          The name of the data window to apply  to  the  data  in
          each  frame  before  computing reflection coefficients.
          If the option is omitted, the parameter  file  is  con-
          sulted,  and  if  no  value is found there, the default
          used is a rectangular window with amplitude one.   Pos-
          sible  window  types include rectangular ("RECT"), Ham-
          ming ("HAMMING"), Hanning ("HANNING"),  cosine  (COS4),
          and  triangular  ("TRIANG");  see  the window(3-ESPSsp)
          manual page.  If the last frame is truncated, the  win-
          dow  is applied to the truncated data (e.g., a triangu-
          lar window is zero at the start and end  of  the  trun-
          cated data).

     ----mmmm _m_e_t_h_o_d[autoc]
          Specifies the spectrum analysis method.  The default is
          the  autocorrelation  method.   Also  available are the
          covariance method ("cov"), Burg method ("burg"),  modi-
          fied  Burg  method ("mburg"), fast modified Burg method
          ("fburg"),   stuctured   covariance    ("strcov"    and
          "strcov1"),  and  vector Burg ("vburg", fast approxima-
          tion to structured covariance.  Of the  two  structured
          covariance methods, the first ("strcov" is considerably
          faster and  better  behaved;  "strcov1"  is  older  but
          included  as  it  may  prove  useful  on  occasion.  If
          "strcov"  is  used,  the  ----cccc  and  ----iiii  options   become
          relevant.   The  ----mmmm option overrides the value that may
          be in the parameter file.  The default applies only  if
          there is no value in the parameter file.

     ----eeee _p_r_e_e_m_p_h_a_s_i_s[0.0]
          Specifies a preemphasis factor to apply  to  the  input
          signal.

     ----cccc _c_o_n_v__t_e_s_t[1e-5]
          Specifies, for the STRCOV method  only  (not  including
          STRCOV1),  a  convergence  test  value.   The lower the
          value, the smaller the change required on  each  itera-
          tion  before  the  estimator  terminates,  and the more
          iterations that normally will result.

     ----iiii _m_a_x__i_t_e_r[20]
          Specifies, for the STRCOV method  only  (not  including
          STRCOV1),  the  maximum  number  of iterations that the
          estimator will run through before terminating.  A warn-
          ing  will  indicate if the estimator terminates because
          max_iter has been exceeded.




        Copyright Entropic Research Laboratory, Inc. 4G4        4






REFCOF(1-ESPS)                              ENTROPIC RESEARCH LAB



     ----oooo _o_r_d_e_r[15]
          Specifies  the  order  (number  of  reflection  coeffi-
          cients),  and  overrides  the  value that may be in the
          parameter file.  The default applies only if  there  is
          no value in the parameter file.

     ----ssss _s_i_n_c__n
          For the AUTOC, STRCOV, and STRCOV1 methods,  the  auto-
          correlation  the autocorrelation function is multiplied
          by the function _s_i_n (_x / _s_i_n_c__n) before  computing  the
          reflection  coefficients.  In the frequency domain this
          has the effect of convolving the spectrum with a boxcar
          function  of  width _f / _s_i_n_c__n, where _f is the sampling
          frequency.  The value of _s_i_n_c__n is recorded in  a  gen-
          eric header item.

     ----dddd    Specifies that the  dc  component  of  each  frame  is
          removed  before the analysis is performed.  DC revmoval
          takes place before windowing.

     ----ZZZZ    If the last frame normally would  overrun  the  stated
          range,  _r_e_f_c_o_f reads past the range to fill up the last
          frame; if the the last would go past the file end,  the
          frame  is  filled  with  zoes. Use of ----ZZZZ, inhibits this
          behavior by processing one less frame.  The  result  is
          that  the  end  of  the  last  frame falls short of the
          stated range.  A common use of ----ZZZZ is to  avoid  getting
          unwanted zeros in training sequences.