This paper presents the first annotated and phonetically transcribed large speech corpora developed for spoken Czech. All corpora were collected during the last two years at the Department of Cybernetics, University of West Bohemia (UWB) in Pilsen. The first two collections are broadcast news, the third corpus is a high-quality read-speech database. This paper describes the collection conditions, annotation and phonetic transcription process related to each corpus. The basic phonetic and lexical characteristics of all corpora will be given and compared mutually.