摘要 |
PURPOSE:To provide the learning method for environment dependent type pho neme HMM which does not require an environment independent type phoneme HMM dictionary while maintaining the advantages of linking learning. CONSTITUTION:For the 1st link in the linking learning, the environment independent type phoneme HMM(Hidden Markov Model) is not used and in a step 5, the state number of a word (or phrase or sentence) HMM is calculated directly from the state number of the environment dependent type phoneme HMM. In a step 6, a dummy word (or phrase or sentence) HMM is prepared for learning and decomposition and preparation in a step 7 and linking learning in a step 9 are repeated to learn the environment dependent type phoneme HMM. |