Combining frame and turn-level information for robust recognition of emotions within speech | Synapse