Orca AL – SRKW

First developed in 2020 by Google Summer of Code students Diego Rodríguez and Kunal Mehta, OrcaAL is an “active learning” tool. It lets you teach a machine learning model to recognize orca signals & observe how its performance evolves.

  1. The machine presents you with audio samples that it finds confusing.
  2. You help label the sounds, collaborating with a “crowd” of other human annotators.
  3. After enough samples are labeled, the machine re-trains itself using the new data, then generates a new list of sounds that confuse it.
  4. Repeat!
Screenshot of the OrcaAL UI. The 3s audio sample contains an orca call in vessel noise.

Open-source code

Implementations

  1. OrcaAL with Orcasound data and a SRKW call model

Presentations & products

Dec 2020: Meridian webinar on “Sound detection and classification with deep learning”

Orcasound & OrcaAL
Talk by Kunal Mehta, introduced by Scott Veirs on Wednesday, December 16, 2020.

Coming soon:
Dec 2020: Acoustical Society of America talk (virtual, pre-recorded, live Q&A)

Collaborators

  • Google Summer of Code students
    • Kunal Mehta (2020)
    • Diego Rodríguez (2020)
  • Orcasound mentors (see also the Orcasound GSoC organization page)
    • Valentina Staneva (University of Washington, eScience Institute)
    • Jesse Lopez (Axiom Data Science)
    • Abhishek Singh (Google Summer of Code 2019 alum)
    • Dan Olsen (North Gulf Oceanic Society)
    • Hannah Meyers (University of Alaska)
    • Val Veirs (Beam Reach)
    • Scott Veirs (Beam Reach)

More info:

Leave a Reply

Your email address will not be published. Required fields are marked *