Spoken language processing for audio archives of lectures and panel discussions

Intelligent archiving systems of lectures and panel discussions based on automatic transcription and indexing are introduced. Our speech recognition system Julius has been improved to deal with these kinds of spontaneous speech. Transcriptions are automatically edited for improving readability, and then key sentences are indexed for each segment based on statistically derived discourse markers and topic words. For panel discussions consisting of multiple speakers, unsupervised speaker indexing is applied beforehand to segment audio into speaker turns. Thus, we realize efficient browsing of these audio archives.