To facilitate browsing of speech archives, we will investigate a new research problem called taxonomy generation for speech archives in this paper. Speech archives are considered difficult to be browsed and navigated. Although the whole transcription of a spoken document might not be well recognized in a normal case, some key terms still can be recognized. In this study we propose an approach to grouping similar key terms extracted from the transcription of a speech archive into clusters and similar clusters into super clusters to form a subject taxonomy for the archive. We will report the potential merits and challenges of the proposed approach.