An implementation of Hidden Markov Model (HMM) with the following features:

• Unsupervised training using Baum-Welch Algorithm.
• Large scale training by using online learning techniques (mini-batch approach).
• Faster training speed by multi-threading.
• Viterbi Algorithm for decoding, supports multi-threading.
• Platform independent: written entirely in Java.
• Simple and stand-alone implementation (easy to modify).
• Supports unicode.
• Optionally uses development data to avoid overfitting (by checking perplexity).