| 摘要 |
The present invention relates to a method for preparing data for speech recognition, the data being provided in the form of a phonetic string, wherein the phonetic string is modelled with a sequence of models by processing the phonetic string comprising at least one character characterwise, the processing of a character comprising:
€¢ checking for at least one phonetic string segment comprising said character if a model exists for said phonetic string segment in a set of rules associating phonetic string segments with models,
€¢ if a model exists for a checked phonetic string segment, assigning said model to a character comprised in said phonetic string segment if no longer model is already assigned to said character,
wherein the models assigned to the characters form a model sequence representing the phonetic string.
|
