Future-proofing deepfake detection by integrating audio, video, and text