contents

[jelletreep]

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)