В предыдущих исследованиях показана целесообразность применения адаптации системы пофонемного распознавания речи к голосу нового для системы диктора. Смысл адаптации заключается в улучшении распознавания нового диктора. Оговорен вопрос подбора наиболее оптимальных адаптационных выборок.
В попередніх дослідженнях показано доцільність застосування адаптації системи пофонемного розпізнавання мови до голосу нового для системи диктора. Сенс адаптації полягає в покращенні розпізнавання нового диктора. Обговорено питання підбору найбільш оптимальних адаптаційних вибірок.
Introduction: There are tasks, when the recognition system must serve to the plenty of users. All these users will be for the system unknown announcers. The system can not good recognize these announcers. So, a new announcer should have a small adaptation set. Thus, if a new announcer pronounce such adaptation set the growth of exactness of recognition will be perceptible. Purpose: The purpose of work is experimental research of the new announcers speech recognition growth after their adaptation to the recognition system depending on the volumes of the adaptation sets. It is necessary also to probe the dependence of the exactness recognition growth from the high-quality composition of the adaptation sets. Results: It is set in the numerous experiments, that exactness of the new announcer speech recognition after adaptation is increased on the average at the increase of adaptation set size. But this process is observed only on the average. Experiments shown that the practically identical on volume adaptation sets can give different exactness growth of the recognition substantially. The results of recognition are conducted on every adaptation set. After that the most effective set is selected. Conclusions: The method of choosing the most effective adaptation sets is offered. This method is empiric and requires a lot of time. For greater efficiency it is necessary to work with the plenty of new announcers and the plenty of the adaptation sets. A conclusion is done about the amount of words, which must be in adaptation sets.
