Automatic Singing Transcription Based on Encoder-decoder Recurrent Neural Networks with a Weakly-supervised Attention Mechanism.
Ryo NishikimiEita NakamuraSatoru FukayamaMasataka GotoKazuyoshi YoshiiPublished in: ICASSP (2019)
Keyphrases
- recurrent neural networks
- weakly supervised
- attention mechanism
- feed forward
- object class
- topic models
- neural network
- semi supervised
- relation extraction
- superpixels
- motion estimation
- artificial neural networks
- named entities
- machine learning
- information extraction
- pairwise
- object recognition
- human computer interaction
- multi modal
- input image
- low level
- saliency map
- multiscale
- artificial intelligence
- information retrieval