針對所觀測到的信號序列，進而去猜測其實際所在狀態位置，如同在音樂領域上從「音符序列」推估「和弦狀態」、或在生物醫學領域裡從「DNA序列」找尋「遺傳特徵狀態」……等不同的應用，若情境符合隱馬可夫模型，即可利用簡易的貝氏定理，簡化在序列中過多的假設，便於求取所感興趣的狀態跳動的轉換機率；但因為隱馬可夫模型並無法得知狀態位置，所以就需要在所看到的序列長度中，嘗試所有狀態的組合情況，以最大概似估計法，找出發生機率最大的狀態序列組合。
然而，許多情境下狀態數量與序列長度，將超出研究人員的想像，故時間成本必然不符合現實經濟效益；但本研究發現到，隱馬可夫模型參數資訊中，狀態的轉移矩陣在自我狀態跳動的機率，可能相較跳動到其他狀態上明顯高出很多，若能有效假設部份序列是相同狀態的跳動，即可減少為求最佳解所需嘗試的組合數量，越精確的判別與符合該假設的情境，除了能讓該研究方法與窮舉法有不相上下的準確率，最大的貢獻是能大大的降低所花費的時間成本。

Based on the observation of signal sequences, our purpose is to predict the real signal states. In our research, we use the hidden Markov model (HMM) to simplify the complicated hypothesis. It means that we can get the state-transition probability of our interested signals by Bayes’ Theorem. However, we can’t get the signal states from the HMM directly. Instead, we have to check all combinations of state sequence based on the length of signals. Then, we use the maximum likelihood estimation method to get the state sequences with maximum state-transition probability.
In our research, we found out that the probability of state-transition to itself is higher than to other states. If we can assume that some of the state-transitions are the same, the combinations of state sequence we have to check will dramatically decrease. It not only can help us to accurately estimate the real signal state but to effectively reduce the experiment time.

[1] G. DAVID FORNEY, JR, “ The Viterbi Algorithm”. Proceedings of the IEEE, 61(3), p. 268-278, March 1973.
[2] Lloyd R. Welch, “Hidden Markov Models and the Baum-Welch Algorithm”. IEEE Information Theory Society Newsletter, 53(4), December 2003.
[3] Larry Moss, “Example of the Baum-Welch Algorithm”. Q520, Spring 2008.
[4] Lawrence R. RABINER, “A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition”. Proceedings of the IEEE, 77(2), p. 257-286, February 1989.
[5] Alexander Sheh, Daniel P.W. Ellis, “Chord Segmentation and Recognition using EM-Trained Hidden Markov Models”. Proc. ISMIR, 2003.
[6] Kyogu Lee, Malcolm Slaney, “Automatic Chord Recognition from Audio Using an HMM with Supervised Learning”. Proc. ISMIR, 2006.
[7] 林千翔, 張嘉惠, 基於特製隱藏式馬可夫模型之中文斷詞研究. 國立中央大學資訊工程學系
[8]University of California Berkeley, California, Sheldon M. Ross, “Introduction to Probability Models- 9th edition”, ACADEMIC PRESS, 2007
[9]台灣科技大學, 楊維寧博士, 「統計學二版」, 新陸書局, 2007