We address the automatic generation of acoustic subword units and an associated pronunciation dictionary for speech recognition. The speech audio is first segmented into phoneme-like units by detecting points at which ...