Background
Research questions
logistics Hard drives with TB of audio data by plane
Lot of data
Data on Yoda
Team and collaboration
Scientific problems:
• Not a lot of vocalizations
• Recordings Sanctuary, how representative is that
Annotations:
• Raven for annotations
• Speeding up annotation
• Synthetic data
Machine learning:
• SVM: feature extraction and selection
• CNN: link to challenge, CNN selection, finetuning
• <>
• Interspeech challenge
• Challenge test dataset is from sanctuary. How representative is this result in “rainforest”
• Finetuning PANNS on training data
Results
• Model performance
• Publish dataset
Lessons learned:
• Collaboration with ML audio expert
• Involvement researcher
• Data management
• Code management (Matlab scripts, repo’s with playground folders, no real git workflow, documentation, package from early stage)
Notes Animal sounds project
Background Research questions logistics Hard drives with TB of audio data by plane Lot of data Data on Yoda
Team and collaboration
Scientific problems: • Not a lot of vocalizations • Recordings Sanctuary, how representative is that
Annotations: • Raven for annotations • Speeding up annotation • Synthetic data
Machine learning: • SVM: feature extraction and selection • CNN: link to challenge, CNN selection, finetuning • <> • Interspeech challenge • Challenge test dataset is from sanctuary. How representative is this result in “rainforest” • Finetuning PANNS on training data
Results • Model performance • Publish dataset
Lessons learned: • Collaboration with ML audio expert • Involvement researcher • Data management • Code management (Matlab scripts, repo’s with playground folders, no real git workflow, documentation, package from early stage)