Spoken language translation from parallel speech audio: Simultaneous interpretation as SLT training data