Source Code for Module feature_vec

  1  #!/usr/bin/env python 
  2   
  3  import os 
  4  import sys 
  5  import math as M 
  6  from gmisclib import Num 
  7  from gmisclib import die 
  8  # from gmisclib import erb_scale 
  9  from gmisclib import avio 
 10  from gmisclib import cache as CC 
 11  import gpkavg 
 12  import gpkimgclass 
 13   
 14  # from gpk_voicing import percep_spec 
 15  # from gpk_voicing import voice_misc 
 16  from gpk_voicing import fv_misc as FVM 
 17  # import irregularity 
 18  # sys.path.insert(0, '%s/tick1/bin' % os.environ['MRImodel']) 
 19   
 20  if 'PYLABDISP' in os.environ: 
 21          if os.environ['PYLABDISP'] == 'pylab': 
 22                  import pylab 
 23          else: 
 24                  import g_pylab as pylab 
 25          PLOT = True 
 26  else: 
 27          PLOT = False 
 28   
 29  ROOT = '/tmp/fv_cache' 
 30   
 31   
 32   
 33  # SpecExp = 0.0 
 34   
 35   
 36   
 37   
 38 -def average(x): 
 39          vv = [] 
 40          ww = [] 
 41          for (v,wt) in x: 
 42                  vv.append(v) 
 43                  ww.append(wt) 
 44          return gpkavg.avg(vv, ww, 0.0)[0] 
 45   
 46 -def median(x): 
 47          vv = [] 
 48          ww = [] 
 49          for (v,wt) in x: 
 50                  vv.append(v) 
 51                  ww.append(wt) 
 52          return gpkavg.avg(vv, ww, 0.499)[0] 
 53   
 54   
 55 -def weight(x): 
 56          sum = 0.0 
 57          for (v,wt) in x: 
 58                  sum += wt**2 
 59          return M.sqrt(sum) 
 60   
 61   
 62 -class matmult_xform(object): 
 63          from gmisclib import chunkio 
 64 -        def __init__(self, filename): 
 65                  self._name = filename 
 66                  self.xform = self.chunkio.datachunk(open(filename, 'r')).read_NumArray() 
 67   
 68 -        def name(self): 
 69                  return self._name 
 70   
 71 -        def describe_xform(self, descr): 
 72                  pass 
 73   
 74 -        def operate(self, o, descr): 
 75                  Num.matrixmultiply(o, self.xform, o) 
 76                  for i in range(o.shape[1]): 
 77                          descr[i] = {'id': 'Mixture%d'% i} 
 78   
 79   
 80   
 81 -def describe_xform(descr, fac): 
 82          nin, nout = fac.shape 
 83          assert len(descr) == nin 
 84          for i in range(nout): 
 85                  norm = M.sqrt(Num.average(fac[:,i]**2)) 
 86                  nf = fac[:,i]/norm 
 87                  tmp = [] 
 88                  fwt = [] 
 89                  wwt = [] 
 90                  edginess = [] 
 91                  voiciness = [] 
 92                  burstiness = [] 
 93                  friciness = [] 
 94                  vness = [] 
 95                  for j in range(nin): 
 96                          dj = descr[j] 
 97                          if dj.has_key('erb'): 
 98                                  fwt.append( (dj['erb'], abs(nf[j])) ) 
 99                          if dj.has_key('width'): 
100                                  wwt.append( (dj['width'], abs(nf[j])) ) 
101                          if 'edge' in dj['type']: 
102                                  edginess.append( abs(nf[j]) ) 
103                          if 'burst' in dj['type']: 
104                                  burstiness.append( abs(nf[j]) ) 
105                          if 'fricative' in dj['type']: 
106                                  friciness.append( abs(nf[j]) ) 
107                          if 'vowel' in dj['type']: 
108                                  vness.append( abs(nf[j]) ) 
109                          if 'haspitch' in dj['type']: 
110                                  voiciness.append( abs(nf[j]) ) 
111                          if abs(nf[j]) > 0.1: 
112                                  tmp.append('%+.2f*%s' % (nf[j], dj['id']) ) 
113                  print '# FV[%d]= %s' % (i, ' '.join(tmp)) 
114                  fb = {'voicing': M.sqrt(Num.sum(Num.array(voiciness)**2)), 
115                          'edges': M.sqrt(Num.sum(Num.array(edginess)**2)), 
116                          'burst': M.sqrt(Num.sum(Num.array(burstiness)**2)), 
117                          'fricative': M.sqrt(Num.sum(Num.array(friciness)**2)), 
118                          'vowel': M.sqrt(Num.sum(Num.array(vness)**2)), 
119                          'meanfreq': average(fwt), 'medfreq': median(fwt), 
120                          'wt_freq': weight(fwt), 
121                          'meanwidth': average(wwt), 'medwidth': median(wwt), 
122                          'wt_width': weight(wwt), 
123                          'channel': i, 'description': ' '.join(tmp), 
124                          'ltype': 'fb' 
125                          } 
126                  print avio.concoct(fb) 
127   
128   
129   
130   
131 -def run(argv): 
132          Lf = 1.5 
133          Elf = 1.5 
134          Nsv = FVM.NSV 
135          # Nsv and LF updated from the 5 April 2007 optimization 
136          # in .../m/ASR/OPT 
137          DT = 0.01       # Seconds.  Default output sampling interval. 
138          Scale = None 
139          arglist = argv[1:] 
140          arglist0 = arglist 
141          column = None 
142          signalfile = None 
143          outfile = None 
144          Cache = None 
145          verbose = 0 
146          kwargs = {} 
147          import gpk_voicing.fvcurrent as fvm 
148          extrahdr = {} 
149          while arglist and arglist[0].startswith('-'): 
150                  arg = arglist.pop(0) 
151                  if arg == '--': 
152                          break 
153                  elif arg == '-DT' or arg == '-dt': 
154                          DT = float(arglist.pop(0)) 
155                  elif arg == '-f': 
156                          signalfile = arglist.pop(0) 
157                  elif arg == '-c' or arg == '-col': 
158                          tmp = arglist.pop(0) 
159                          try: 
160                                  column = int( tmp ) 
161                          except ValueError: 
162                                  column = tmp 
163                          die.note("signalfile", signalfile) 
164                  elif arg == '-o': 
165                          outfile = arglist.pop(0) 
166                  elif arg == '-v': 
167                          verbose += 1 
168                  elif arg == '-cache': 
169                          Cache = True 
170                  elif arg == '-Nsv': 
171                          FVM.NSV = float(arglist.pop(0)) 
172                  # elif arg == '-SpecExp': 
173                          # SpecExp = float(arglist.pop(0)) 
174                  elif arg == '-sfv': 
175                          import gpk_voicing.fvsimple as fvm 
176                          kwargs['do_voicing'] = 1 
177                          kwargs['do_dissonance'] = True 
178                          kwargs['ELF'] = Elf 
179                  elif arg == '-ssfv': 
180                          import gpk_voicing.fvss as fvm 
181                          kwargs['do_dissonance'] = False 
182                          kwargs['do_voicing'] = 1 
183                          kwargs['ELF'] = Elf 
184                  elif arg == '-specfv': 
185                          import gpk_voicing.fvss as fvm 
186                          kwargs['do_dissonance'] = False 
187                          kwargs['do_voicing'] = 0 
188                          kwargs['PlompBouman'] = False 
189                          kwargs['ELF'] = Elf 
190                  elif arg == '-fv20071030': 
191                          import gpk_voicing.fv20071030 as fvm 
192                          kwargs['do_dissonance'] = True 
193                          kwargs['do_voicing'] = 1 
194                          kwargs['ELF'] = Elf 
195                  elif arg == '-fv20080325': 
196                          import gpk_voicing.fv20080325 as fvm 
197                          kwargs['do_dissonance'] = True 
198                          kwargs['do_voicing'] = 1 
199                          kwargs['ELF'] = Elf 
200                  elif arg == '-fv200807jyuan': 
201                          import gpk_voicing.fv200807jyuan as fvm 
202                          kwargs['do_voicing'] = 1 
203                          kwargs['do_dissonance'] = True 
204                          kwargs['ELF'] = Elf 
205                  elif arg == '-fv200811align': 
206                          import gpk_voicing.fv200811align as fvm 
207                          kwargs['do_voicing'] = 2 
208                          kwargs['do_dissonance'] = True 
209                          kwargs['ELF'] = Elf 
210                  elif arg == '-fv200812align': 
211                          import gpk_voicing.fv200812align as fvm 
212                          kwargs['do_voicing'] = 2 
213                          kwargs['do_dissonance'] = True 
214                          Lf = 1.0 
215                          Elf = 1.0 
216                          kwargs['ELF'] = Elf 
217                  elif arg == '-fv200908opt': 
218                          import gpk_voicing.fv200908opt as fvm 
219                          kwargs['do_voicing'] = 1 
220                          kwargs['do_dissonance'] = False 
221                          kwargs['ELF'] = Elf 
222                  elif arg.startswith('-fv200909opt'): 
223                          import gpk_voicing.fv200909opt as fvm 
224                          kwargs['do_voicing'] = 1 
225                          kwargs['do_dissonance'] = False 
226                          Nsv, Lf, Elf = fvm.Scale.get_many('nsv', 'lf,lf', 'lf,elf') 
227                          if arg.endswith('S'): 
228                                  Scale = fvm.Scale 
229                          kwargs['ELF'] = Elf 
230                  elif arg.startswith('-fv201003opt'): 
231                          import gpk_voicing.fv201003opt as fvm 
232                          kwargs['do_voicing'] = 1 
233                          kwargs['do_dissonance'] = False 
234                          Nsv, Lf, Elf = fvm.Scale.get_many('nsv', 'lf,lf', 'lf,elf') 
235                          if arg.endswith('S'): 
236                                  Scale = fvm.Scale 
237                          kwargs['ELF'] = Elf 
238                  elif arg.startswith('-fv201102'): 
239                          import gpk_voicing.fv201102 as fvm 
240                          Nsv, Lf = fvm.Scale.get_many('norm,nsv', 'lf,lf') 
241                          kwargs['tdif'], kwargs['tdif2'] = fvm.Scale.get_many('norm,tdif', 'norm,tdif2') 
242                          kwargs['fdif2'], kwargs['FORMANT_HIGH'] = fvm.Scale.get_many('norm,fdif2', 'FORMANT_HIGH') 
243                          if arg.endswith('S'): 
244                                  Scale = fvm.Scale 
245                  elif arg == '-LF': 
246                          Lf = float(arglist.pop(0)) 
247                  elif arg == '-ELF': 
248                          Elf = float(arglist.pop(0)) 
249                  elif arg == '-novoicing': 
250                          kwargs['do_voicing'] = 0 
251                  elif arg == '-Xform': 
252                          Scale = matmult_xform(arglist.pop(0)) 
253                  elif arg == '-write': 
254                          extrahdr = dict( [q.strip().split('=', 1) 
255                                          for q in arglist.pop(0).split(';') ] 
256                                          ) 
257                  else: 
258                          die.info("Unrecognized flag: %s" % arg) 
259                          print __doc__ 
260                          die.exit(1) 
261          if arglist and signalfile is None: 
262                  signalfile = arglist.pop(0) 
263          if arglist and outfile is None: 
264                  outfile = arglist.pop(0) 
265          elif outfile is None: 
266                  outfile = 'fv.dat' 
267          if signalfile is None: 
268                  die.info("No signal file specified.") 
269                  print __doc__ 
270                  die.exit(1) 
271          if arglist: 
272                  die.info('Extra arguments!') 
273                  print __doc__ 
274                  die.exit(1) 
275          signal = gpkimgclass.read(signalfile) 
276          if column is None and signal.d.shape[1]==1: 
277                  column = 0 
278          elif column is None: 
279                  die.die("There are %d channels in %s, so you must specify one with '-c'." % (signal.d.shape[1], signalfile)) 
280          try: 
281                  signal.column(column) 
282          except KeyError: 
283                  die.info("File has %d columns" % signal.n[1]) 
284                  die.die("Bad column: %s" % str(column)) 
285   
286          if Cache is not None: 
287                  kwargs['cache_info'] = CC.cache_info(root=ROOT, file=signalfile, info=(column,)) 
288          o, descr, DTx, tshift = fvm.feature_vec(signal.column(column), signal.dt(), DT, 
289                                          Nsv=Nsv, LF=Lf, **kwargs 
290                                          ) 
291          o = Num.transpose(o) 
292          hdr = signal.hdr.copy() 
293          if Scale is not None: 
294                  Scale.operate(o, descr) 
295                  hdr['XFORM'] = Scale.name() 
296                  if verbose: 
297                          Scale.describe_xform(descr) 
298   
299          hdr['program'] = argv[0] 
300          hdr['ARGV'] = arglist0 
301          hdr['input_file'] = signalfile 
302          hdr['column'] = column 
303          hdr['CDELT2'] = DTx 
304          hdr['CRPIX2'] = 1 
305          hdr['CRVAL2'] = signal.start() + tshift 
306          hdr['CDELT1'] = 1 
307          hdr['BITPIX'] = -32 
308          hdr['DataSamplingFreq'] = 1.0/signal.dt() 
309          hdr['Nsv'] = FVM.NSV 
310          hdr['LF'] = Lf 
311          hdr['ELF'] = Elf 
312          hdr.update( extrahdr ) 
313   
314          if PLOT: 
315                  pylab.matshow(Num.transpose(o)) 
316                  pylab.show() 
317          assert o.shape[1] == len(descr), "Length mismatch data=%s descr=%d" % ( 
318                                                  o.shape, len(descr) 
319                                                  ) 
320          for (i,d) in enumerate(descr): 
321                  hdr['TTYPE%d' % (i+1)] = d['id'] 
322                  hdr['F_INFO%d' % (i+1)] = avio.concoct(d) 
323                  hdr['RMS%d' % (i+1)] = M.sqrt(Num.average(o[:,i]**2))  
324          gpkimgclass.gpk_img(hdr, o).write(outfile) 
325   
326   
327   
328  if __name__ == '__main__': 
329          # try: 
330                  # import psyco 
331                  # psyco.full() 
332          # except ImportError: 
333                  # pass 
334   
335          run(sys.argv) 
336