Project Plan

Meeting with Ole Winther about Project

Approach to project

In his opinion a difficult project, take start in similar projects. His immediate ideas for the project was to use CNN or LSTM to classify the images. He also noted, that the classification in itself might be difficult, and the Generation idea raised an eyebrow.

Papers

He did not mention any papers, but told us to research, and that we should not start from scratch.

Approach

Start with small data set and small model, and increase model complexity and size of data set.

Formalia of project / Expectations

What is evaluated in the project is our approach to the problem, and especially how do we construct and go about the project. Talking points were:

• If a result does not make sense, how do we check what goes wrong
  • Listening to the sound, does it make sense for us, that the model might confuse it
  • Are the sounds similar or far apart in our hearing opinion
  • Are the sounds similar or far apart in the spectrogram
  • Are we looking at the correct stuff
• Does it make sense to use the filters we use for the CNN
  • Using the spectrogram, the filters might see one

Structure of report

• Problem statement / Goal
  • Why is this relevant. He talked about species preservation or non-invasive investigations of species.
• State of the art
  • This will also be what the starting point of our project. What did others do, and maybe we can use what they already have.
• Data
  • Bla bla animal sounds
• Model
• Evaluation
  • Here we should not only evaluate how well our model performs but also our approach to the project. Maybe reflect over what worked well, and what didn't as well
• Conclusion