Multi-channelled content-based encoding synthesize one-to-one models