CNN Based Two-stage Multi-resolution End-to-end Model for Singing Melody Extraction.