Skip to content
No description, website, or topics provided.
Branch: master
Clone or download
1
Latest commit 2141fe6 Sep 4, 2017
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
obama_data intial commit Sep 2, 2017
README.md Update README.md Sep 4, 2017
run.py
util.py

README.md

This is research-code for Synthesizing Obama: Learning Lip Sync from Audio.
Supasorn Suwajanakorn, Steven M. Seitz, Ira Kemelmacher-Shlizerman
SIGGRAPH 2017

Code tested using tensorflow 0.11.0 Please see Supasorn's website for the overview.

To generate MFCC, first normalize the input audio using https://github.com/slhck/ffmpeg-normalize. Then use Sphinx III's snippet by David Huggins-Daines with a modified routine that saves log energy and timestamps:

def sig2s2mfc_energy(self, sig, dn):
  nfr = int(len(sig) / self.fshift + 1)

  mfcc = numpy.zeros((nfr, self.ncep + 2), 'd')
  fr = 0
  while fr < nfr:
    start = int(round(fr * self.fshift))
    end = min(len(sig), start + self.wlen)
    frame = sig[start:end]
    if len(frame) < self.wlen:
      frame = numpy.resize(frame,self.wlen)
      frame[self.wlen:] = 0
    mfcc[fr,:-2] = self.frame2s2mfc(frame)
    mfcc[fr, -2] = math.log(1 + np.mean(np.power(frame.astype(float), 2)))
    mid = 0.5 * (start + end - 1)
    mfcc[fr, -1] = mid / self.samprate

    fr = fr + 1
  return mfcc
You can’t perform that action at this time.