Semi-automatic processing and annotation of meeting audio recordings

The main theme of this article is creating a test database consisting of meeting audio recordings. The article describes creation and experimental evaluation of proposed database. The principle of automatic speech recognition is briefly described in the introduction. The basic scheme of the functioning of the Kaldi Speech Recognition System is described below. The main part of the thesis is the presentation of the results of the experimental evaluation of the Kaldi system by the proposed testing database. In the end the possibility of improving the results of the Kaldi system by adding automatically recognized data to the system training process is described.