Training general-purpose audio tagging networks with noisy labels and iterative self-verification.
Matthias DorferGerhard WidmerPublished in: DCASE (2018)
Keyphrases
- general purpose
- training set
- training examples
- special purpose
- application specific
- complex networks
- training process
- training data
- domain specific
- social networks
- active learning
- multimedia
- metadata
- programming language
- noisy data
- test set
- low quality
- audio signals
- signal processing
- recurrent networks
- model checking
- labelled data
- echo state networks
- training samples
- supervised learning
- annotation effort
- tag recommendation
- multi label
- label noise
- audio stream
- noisy environments
- community structure
- network structure
- neural network
- missing data
- pairwise
- video sequences
- learning algorithm