Automatic indexing of TV and radio speech data requires robust components for both speech recognition and spoken document retrieval. Due to the high topic variability and the resulting large vocabularies, classic word-based approaches have to cope with a high number of out-of-vocabulary words. This talk presents a phonetic approach to open vocabulary indexing based on syllable decoding and retrieval. Current experimental results are presented, followed by a demonstration of the Fraunhofer IAIS AudioMining system for spoken term detection.