Pronunciation error detection device, pronunciation error detection method, program

摘要

Provided is a pronunciation error detection device capable of tracking even when a correct sentence is not essential and an erroneous recognition such as a reading error occurs. A voice recognition unit that performs voice recognition of voice data based on a non-native speaker voice recognition model and outputs the voice recognition result, its reliability, and its time information, and a voice whose reliability is higher than a preset threshold. Low-constrained grammar native speaker voice for the reliability judgment unit that outputs the recognition result and the corresponding time information as the speech recognition result after the judgment and the time information after the judgment, and the voice data of the section specified by the time information after the judgment. The reliability of each speech element of the speech recognition result using the recognition model is higher than the reliability of the corresponding speech element of the speech recognition result by the native speaker acoustic model under the grammatical constraint that the speech recognition result is the correct answer after the judgment. In this case, it includes a speech error detection unit that outputs the phonetic element as a speech error.