Tail distribution modelling using the richter and power exponential distributions
نویسندگان
چکیده
The vast majority of HMM–based speech recognition systems use Gaussian mixture models as the state distribution model. The use of these distributions is motivated more by ease of training, decoding and the fact that a sufficient number of Gaussian components may be used to approximate any distribution, than some underlying aspect of the data being modelled. If distributions were selected that better modelled the observed data, fewer components should be required and recognition accuracy should improve. This paper examines two distributions for improving the modelling of the tails of the densities. The first distribution, the Richter distribution, fits within the general framework of Gaussian component tying, but has some attractive attributes for decoding. The second distribution, the power exponential, does not fit within a tying framework. Despite gains in likelihood, indicating that the Gaussian components are sub–optimal in a likelihood sense, only small gains in recognition performance were observed on a large vocabulary speech recognition task.
منابع مشابه
On Bivariate Generalized Exponential-Power Series Class of Distributions
In this paper, we introduce a new class of bivariate distributions by compounding the bivariate generalized exponential and power-series distributions. This new class contains the bivariate generalized exponential-Poisson, bivariate generalized exponential-logarithmic, bivariate generalized exponential-binomial and bivariate generalized exponential-negative binomial distributions as specia...
متن کاملOn Exponential Power Distribution And Poultry Feeds Data: A Case Study
Abstract. In this paper, we propose to study a generalized form of the exponential power distribution which contains others in the literature as special cases. This unifying exponential power distribution is characterized by a parameter ω and a function h(ω) which regulates the tail behavior of the distribution, thus making it more flexible and suitable for modeling than the usual normal di...
متن کاملProbability Distribution Fitting to Maternal Mortality in Nigeria.
The consequences of Maternal Mortality (MM) cannot be overemphasized. It inhibits population growth resulting into loss of lives among others. This work tends to obtain the maternal mortality rates (MMR) in Nigeria, identify some fitted distributions to MMR and determine which of the distributions best fits the data. A comprehensive Exploratory Data Analysis (EDA) was carried on MM and the MMRs...
متن کاملHigher moments portfolio Optimization with unequal weights based on Generalized Capital Asset pricing model with independent and identically asymmetric Power Distribution
The main criterion in investment decisions is to maximize the investors utility. Traditional capital asset pricing models cannot be used when asset returns do not follow a normal distribution. For this reason, we use capital asset pricing model with independent and identically asymmetric power distributed (CAPM-IIAPD) and capital asset pricing model with asymmetric independent and identically a...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 1999