PROBLEM TO BE SOLVED: To provide a spoken language identification device capable of highly reliably identifying a language of speech on the basis of speech data.SOLUTION: A learning device comprises a storage device for language-labeled speech data; a block feature generation part 180 for extracting, from the speech data, a series of speech features with a prescribed time length and a prescribed shift length; a codebook calculation part 184 for generating a codebook on the basis of the extracted series of speech features; a language phoneme feature vector calculation part 186 for obtaining, from the codebook, representative vectors closest to speech features included in the series of speech features obtained from the speech data with respect to each of the plural pieces of speech data and generating language-labeled speech language features of the speech data on the basis of a distribution of the representative vectors; and an SVM learning part 190 for generating an SVM used for estimating a language on the basis of the speech language features with the speech language features used as learning data.
展开▼