a soft segment modeling approach for duration modeling in phoneme recognition systems

the geometric distribution of states duration is one of the main performance limiting assumptions of hidden markov modeling of speech signals. stochastic segment models, generally, and segmental hmm, specifically, overcome this deficiency partly at the cost of more complexity in both training and recognition phases. in this paper, a new duration modeling approach is presented. the main idea of the model is to consider the effect of adjacent segments on the probability density function estimation and evaluation of each acoustic segment. this idea not only makes the model robust against segmentation errors, but also it models gradual change from one segment to the next one with a minimum set of parameters. the proposed idea is analytically formulated and tested on a timit based context independent phoneme classification system. during the test procedure, the phoneme classification of different phoneme classes was performed by applying various proposed recognition algorithms. the system was optimized and the results have been compared with a continuous density hidden markov model (cdhmm) with similar computational complexity. the results show slight improvement in phoneme recognition rate in comparison with standard continuous density hidden markov model. this indicates improved compatibility of the proposed model with the speech nature.