The proposed dissertation deals with a subset of techniques for extracting meaningful information from speech: voice activity detection, transcription, keyword spotting, speaker recognition, language recognition and other possible modalities. It includes investigation into relevant signal processing and machine learning, and experimentation on standard speech data-sets. The topic is related to several projects running in the BUT Speech@FIT group, see http://speech.fit.vutbr.cz/projects. The candidate should have strong background in mathematics, linear algebra and statistics, and experience in one or more of the following disciplines: signal processing, speech signal processing, machine learning, natural language processing, data-mining. He/she should be experienced with usual scientific programming and scripting languages (C, Matlab, Python). Experience with at least one of machine learning/speech toolkits (Theano, Keras, PyTorch, CNTK, Chainer, KALDI, HTK) is a plus. As the the group is international, a good working knowledge of English is required.