Speech recognition system with maximum entropy language models

摘要

The invention relates to a method of setting a free parameter $\mathop{\mathrm{image}}{\mathrm{\λ}}_{\mathrm{\α}}^{\mathrm{ortho}}$;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 $\mathop{\mathrm{image}}{\mathrm{\λ}}_{\mathrm{\α}}^{\mathrm{ortho}}$;that has a number of possible interpretations. This object is achieved in accordance with the invention in that $\mathop{\mathrm{image}}{\mathrm{\λ}}_{\mathrm{\α}}^{\mathrm{ortho}}$;is calculated as follows: $\mathop{\mathrm{image}}\begin{array}{c}{\mathrm{\λ}}_{\mathrm{\α}}^{\mathrm{ortho}}=\log \⁡\left(\frac{m_{\mathrm{\α}}^{\mathrm{ortho},\mathrm{mod}}}{{\mathrm{Nenner}}_{\mathrm{\α}}}\right)\⁢\text{\ }\⁢\mathrm{with}\\ m_{\mathrm{\α}}^{\mathrm{ortho},\mathrm{mod}}=\underset{\mathrm{\β}\∈A_i}{\∑}\⁢m_{\mathrm{\β}}^{\mathrm{ortho}}\⁢\text{\ }\⁢\mathrm{and}\⁢\text{\ }\⁢{\mathrm{denominator}}_{\mathrm{\α}}=\underset{\mathrm{\β}\∈\mathrm{Ai}}{\∑}\⁢\exp \⁡\left(-{\mathrm{\λ}}_{\mathrm{\β}}^{\mathrm{ortho}}\right)\·M_{\mathrm{\β}}^{\mathrm{ortho}}.\end{array}$