• See lispeaker_labelfiles.mlf and volunteer_labelfiles.mlf.

• They are structured as follows:

  1. path to a video file.
  2. on each line: startTime <space> endTime <space> phoneme
  3. .to signify the end of a video file

• You can extract phoneme files per video using extractTCDTIMITaudio. It can also copy the wav files. You can then fix the headers and replace the phonemes by audioSR/transform.py

• These labelfiles have been modified slightly from the original TCDTIMIT files:

  1. the path of the videos has to be replaced by wherever you've downloaded and extracted TCDTIMIT (see downloadTCDTIMIT for a script to do that)
  2. in a few videos, a phoneme was repeated twice after eachother. It should be one interval until the next (different) phoneme starts