Unified training of WFSA through a generic interface

This paper describes a unified approach to the training of weighted finite-state automata (WFSA) that is based on a generic interface. Regardless of their internal structure, any automata implementing a simple interface can be managed by the system, not only for decoding but also for training purposes. This novel approach drastically simplifies the effort to incorporate new formalisms into a pattern recognition engine. This methodology has been integrated into Sautrela, a highly modular and pluggable open source package for generic purpose signal processing, focused on speech recognition.