Source Code for Module gmisclib.MFCCFile

  1  import struct 
  2  import sys 
  3   
  4  import numpy 
  5   
  6   
  7  # Class for loading mfccfiles in the C_SDK format. Supports byteswapping. 
  8  # Questions: Tor Andre Myrvoll, myrvoll@research.bell-labs.com 
  9   
 10 -class MFCCFile: 
 11      # This class loads a mfcc file in LASR format and gives access to 
 12      # mfcc vectors indexed by file and position 
 13 -    def __init__(self,filename,DEBUG = 0,BYTEORDER='@'): 
 14          self.DEBUG = DEBUG 
 15          self.BYTEORDER = BYTEORDER 
 16          self.FileName = filename 
 17          long_s = 4 
 18          float_s = 4  
 19   
 20          # See if we should byteswap 
 21          if not struct.pack('%sL' % (self.BYTEORDER),1) == struct.pack('L',1): 
 22              BYTESWAP = 1 
 23          else: 
 24              BYTESWAP = 0 
 25   
 26          fp = open(filename,'r') 
 27          if DEBUG: 
 28              sys.stderr.write("Opening file %s\n" % (filename)) 
 29               
 30   
 31   
 32          # Store the number of sentences in this file 
 33          str = fp.read(long_s) 
 34          self.numSent = struct.unpack('%sL' % (self.BYTEORDER),str)[0] 
 35          if DEBUG: 
 36              sys.stderr.write("    Number of sentences: %d\n" % (self.numSent)) 
 37   
 38          # Store the feature vector dimension 
 39          str = fp.read(long_s) 
 40          self.vecSize = struct.unpack('%sL' % (self.BYTEORDER),str)[0] 
 41          if DEBUG: 
 42              sys.stderr.write("    Vector size: %d\n" % (self.vecSize)) 
 43               
 44   
 45          # Load an array of sentence lengths into a list 
 46          self.sentLength = [] 
 47          for sent in range(self.numSent): 
 48              str = fp.read(long_s) 
 49              length = struct.unpack('%sL' % (self.BYTEORDER),str)[0] 
 50              self.sentLength.append(length) 
 51          if DEBUG: 
 52              sys.stderr.write("\n") 
 53               
 54   
 55          # Now load the entire file and store the vectors in lists 
 56          self.sentList = [] 
 57          sentnum = 0 
 58          index = 0 
 59          veclength = self.vecSize*float_s 
 60          if DEBUG: 
 61              sys.stderr.write("    Loading sentences:") 
 62               
 63   
 64          str = fp.read() 
 65          for sent in range(self.numSent): 
 66              if DEBUG: 
 67                  sys.stderr.write(" %d" % (sent)) 
 68              sentdata = [] 
 69              for vec in range(self.sentLength[sent]): 
 70                  mfccvec = numpy.fromstring(str[index:index+veclength],'f') 
 71                  if BYTESWAP: 
 72                      mfccvec = mfccvec.byteswapped() 
 73                  sentdata.append(mfccvec) 
 74                  index += veclength 
 75                   
 76              self.sentList.append(sentdata) 
 77   
 78          # We're done. Close the file 
 79          fp.close() 
 80          if DEBUG: 
 81              sys.stderr.write("\n") 
 82   
 83 -    def Save(self,filename): 
 84          fh = open(filename,'w') 
 85          # First we write the header 
 86          str = struct.pack('LL',self.numSent,self.vecSize) 
 87          fh.write(str) 
 88   
 89          # Now the array of sentence lengths 
 90          for length in self.sentLength: 
 91              str = struct.pack('L',length) 
 92              fh.write(str) 
 93   
 94          # Finally, we save the cepstral vectors themselves 
 95          for utt in self.sentList: 
 96              for vec in utt: 
 97                  str = vec.tostring() 
 98                  fh.write(str) 
 99   
100          fh.close() 
101               
102 -    def ReturnSent(self,sentnum): 
103          # Returns a list containing the entire sentence 
104          return self.sentList[sentnum] 
105       
106   
107 -    def ReturnVector(self,sentnum,vecnum): 
108          # Returns vector i of utterance j 
109          return self.sentList[sentnum][vecnum] 
110