SCARF is a tool kit for recognizing speech with segmental conditional random fields. It supports a broad spectrum of research into segmental features and detection-based speech recognition, and it is intended to support core academic research in this area.