Toward an Automatically Generated Soundtrack from Low-level Cross-modal Correlations for Automotive Scenarios