1. Introduction
The goal of phoneme recognition is to uncover underlying phoneme sequences from speech utterances. Although, it is widely accepted that syntactic and semantic constraints play a major role in automatic speech recognition. However, such constraints are absent in some speech recognition tasks such as keyword searching, segment graph construction for segment-based speech recognition, as well as on-the-fly dictionary manipulations. The performances of these tasks solely rely on acoustic characteristics of the signal. Appropriate acoustic models are the most essential factor affecting the recognition performance. Most state-of-the-art speech recognition systems utilize Hidden Markov Models (HMM) as their acoustic models. HMM Topology Estimation is about searching for an appropriate combination of the number of HMM states together with their connectivity. Estimation is usually performed by choosing an objective function, the measure of merit of each topology, as well as a topology generation method which controls how the topology should evolve itself in the process of searching for the best configuration.