The SIMPLE4ALL project created speech synthesis technology that learns from data with little or no expert supervision and continually improves itself, simply by being used.
ALISA
ALISA uses a two step approach for the task of aligning speech with imperfect transcripts: 1) sentence-level speech segmentation and 2) sentence-level speech and text alignment. Both processes are fully automated and require as little as 10 minutes of manually labelled speech: inter-sentence silence segments for the segmentation, and orthographic transcripts of these sentences for the aligner.
The tool can be applied to any language with an alphabetic writing system and can align up to 75% of the original data with a sentence error rate of less then 8% and a word error rate of less than 1%.
compatibility: Linux/OS X
