ИСТИНА |
Войти в систему Регистрация |
|
Интеллектуальная Система Тематического Исследования НАукометрических данных |
||
Results of the forced alignment may depend in unexpected ways on the transcription, incl.improved results with a “corrupted” transcription. For the given use case (forced alignment with available chunked transcription), it’s worth trying to train a general model less sensitive to language-specific properties of phones — e.g. by using classes based on similarities of acoustic transitions.