مشاهده مشخصات مقاله
Farsi phoneme duration modelling using multivariate adaptive regression splines (MARS)
|
Authors |
-
Masoomeh Bahreini
-
Mohammad Mehdi Homayounpour
|
|
Conference |
دوازدهمین کنفرانس بینالمللی سالانه انجمن کامپیوتر ایران |
|
Abstract |
In this paper, the application of “Multivariate Adaptive Regression Splines” (MARS) to the problem of segmental
(phonemic) duration modelling in Farsi text-to-speech systems is presented. Segmental duration is influenced by a
number of contextual factors such as segment identity, stress, position of a target segment within a syllable, word, and
phrase. These factors interact with each other and a good model of segment duration should account for the problem of
factor's interaction. Databases of speech data often encounters with sparse data problem. MARS is a technique to
estimate general functions of high-dimensional arguments given sparse data, which automatically selects the parameters
and the structure of the model based on available data and deals with the problem of interaction between factors.
Besides highly accurate prediction, a MARS model also allows interpretation of its structure. Using MARS method for
Farsi segmental duration modeling yields a correlation coefficient of 86.50 between observed and predicted durations
for training data of and a correlation coefficient of 80.83 between observed and predicted durations for testing data. The
performance of MARS model was also compared to Multi-Layer Perceptron (MLP) neural network. MLP neural
netwok was trained using an error Back Propagation algorithm. Using MLP neural network for segmental duration
medeling of Farsi language leads to a model with a correlation coefficient between observed and predicted durations of
84.86 for training data and 80.97 for testing data. |
|
قیمت |
-
برای اعضای سایت : 100,000 Rial
-
برای دانشجویان عضو انجمن : 20,000 Rial
-
برای اعضای عادی انجمن : 40,000 Rial
|
خرید مقاله
|
|
.gif){kind=icon}