S4D

Authors:Sylvain MEIGNIER & Anthony LACHER Version:0.1.14 of 2015/07/12

This package is the core of the S4D toolkit.

While developing S4D, we tried to keep in mind two

targets:

1. limit the number of classes to allow better readability

2. make S4D compatible with existing tools

To reach this target, we have created one Main Classe Diar to manage a

diarization annotation.

Diar can be used with a number of tools to perform segmentation and

clustering.

Front-end acoustic feature extraction is perform using SIDEKIT

FeatureServer as well as statistic and model computation

(StatServer and Mixture)

DER scores is handled in two package scoring for single and cross-show.

The frontend package include a number of tools to extract and normalize

• Diar
• Gaussian Model
• Segmentation
• Clustering
• Viterbi
• Utils
• Scoring