TRAP-TANDEM: data-driven extraction of temporal features from speech | IEEE Conference Publication | IEEE Xplore