Speaker adaptation device and manner of acoustic model for speech recognition

摘要

To obtain a device, in which finer speaker adaptive is made possible, by conducting speaker clustering for every fine group of phoneme and generating and selecting an adequate speaker cluster. SOLUTION: The device is provided with a phoneme grouping means 21, which finely conduct phoneme grouping in accordance with data and a speaker clustering means 22, which conducts individual clustering and selecting for each phoneme group and a cluster selecting means 23 and speaker adaptive is realized by speaker cluster selection for individual phoneme. Note that speaker clustering is conducted for every fine phoneme group in accordance with the uttering of an adaptive speaker and a cluster is selected. As a result, learning speaker uttering data, that are more closer to an adaptive speaker and is finely collected.