SPEECH RECOGNITION SYSTEM WITH MAXIMUM ENTROPY LANGUAGE MODELS

摘要

The invention relates to a method of setting a free parameter 1 ${\mathrm{lambda}}_{\mathrm{alpha}}^{\mathrm{ortho}}$ of an attribute in a maximum-entropy speech model, which free parameter could not be set previously with the help of a training algorithm. It is an object of the invention to provide a speech recognition system 100, a training device 10 and a method of setting such a parameter 2 ${\mathrm{lambda}}_{\mathrm{alpha}}^{\mathrm{ortho}}$ that has a number of possible interpretations. This object is achieved in accordance with the invention in that 3 ${\mathrm{lambda}}_{\mathrm{alpha}}^{\mathrm{ortho}}$ is calculated as follows: 4 $\begin{array}{c}{\mathrm{lambda}}_{\mathrm{alpha}}^{\mathrm{ortho}}=\log \left(\frac{m_{\mathrm{alpha}}^{\mathrm{ortho},\mathrm{mod}}}{{\mathrm{Nenner}}_{\mathrm{alpha}}}\right)itit\mathrm{with}\\ m_{\mathrm{alpha}}^{\mathrm{ortho},\mathrm{mod}}=\sum _{\mathrm{beta}\in A_i}it{m}_{\mathrm{beta}}^{\mathrm{ortho}}itit\mathrm{and}itit{\mathrm{denominator}}_{\mathrm{alpha}}=\sum _{\mathrm{beta}\in \mathrm{Ai}}\end{array}$