Dexter

Dexter is a complete diarisation system based on existing open source toolkits. the system extracts the most relevant features from a non-transcribed speech file and then diarises the file, by detecting who is speaking when. all the process is based on unsupervised techniques, not needing labelled filed for training or fine tuning speech models. Dexter includes a style diariser which is a subset of the tools specialised in detecting narrator and non-narrator styles in a mono-speaker audiobook. If the book has been transcribed and aligned in a sentence-by-sentence basis, the style diarisation process can be improved, although this is not mandatory.

Compatability: Linux / Windows

[Download]