摘要 |
<p>A device and method for creating a standard pattern resisting against speech fluctuations or for adapting the speaker, by extracting the feature vector time series of not one kind but a plurality of kinds of feature extracting intervals / starting positions, from a speech for creating a standard pattern or for learning the speaker, to increase the amount of learning data equivalently. By extracting a plurality of feature vector time series of a plurality of feature extracting interval and starting positions from one speech and by using the extracted time series for creating the standard pattern and for learning the speaker, thereby is obtaining an effect nearly equivalent to that obtained by collecting many speeches.</p> |