Orcasound app: live.orcasound.net

An open-source solution for streaming live ocean sound
to citizen scientists and cloud-based algorithms

  • Scott Veirs, Beam Reach, Seattle, WA
  • Val Veirs, Beam Reach, Friday Harbor, WA
  • Paul Cretu, TeamFind, San Francisco, WA
  • Steve Hicks, Einhorn Engineering, Encinitas, CA
  • Skander Mzali, TeamFind, San Francisco, CA
  • Tyler Crisafulli, Google, San Francisco, CA

06 Nov 2018: ASA meeting
Victoria, B.C., Canada

Slides: orcasound.net/data/talks/

Gmail: sveirs, vveirs

Code & collaboration:
Github | Slack | Trello

Orcasound hydrophone network

A cooperative of education/outreach NGOs, researchers, & citizen scientists

  • 15 years of live-streaming to listening community
  • Up to 5 cabled hydrophone nodes in the U.S.
  • 13 NGO members as of fall, 2018
  • 4 educational nodes with physical exhibits using live streams & recordings
  • 10s of dedicated, year-round listeners
  • 100s of listeners during "viral" events

You can join Orcasound, too!
Just e-sign the MOA and heed the
Creative Commons (share-alike, attribution, non-commercial) license.

Orcasound's (historic) automated detection & classification

Admiralty Inlet study (unpublished) -- Of 22 SRKW transits during local daytime,
humans detect 45%, Wholistener 64%, combined 77%.

Complaints lead to new goals & vision

Complaints:

  1. Device/player incompatible with stream (e.g. Shoutcast, Icecast)
  2. Limited listening "seats"
  3. No continuous archived recording (only short clips from algorithms with amplitude or tonal triggers)
  4. Hard drive and hydrophone maintenance

Goals:

  1. Make it easy for unlimited citizen scientists to listen for whales.
  2. Reduce cost & complexity of adding hydrophone nodes.
  3. Stream and archive data in the cloud to promote synergy between human and machine learning.

Orcasound 2.0: new hard/software & a web app in 2018

Vision:
  • open source software
  • open data access
  • real-time engagement of citizen scientists
  • cloud-computing

Thanks to 2017 Kickstarter backers, plus key web app developers/designers to date:
Paul Cretu, Skander Mzali, Steve Hicks, Tyler Crisafulli, Nóra Mészáros, Liam Reese, and participants in the October 2018 Civic Hackathon

Public release of Orcasound app Nov 1, 2018!
live.orcasound.net

New node hardware

Can we deploy a hydrophone node for less than $1000 (U.S.)?!

Basic ingredients (and costs):

  • ($ 50) DIY hydrophone stand
  • ($300 x2) Hydrophones -- dual for redundancy, stereo listening, and bearing determination
  • ($125) Pisound board -- 48/96/192 kHz per channel, stereo 24-bit ADC
  • ($ 35) Raspberry Pi 3b+ -- single board computer (1.4GHz 64-bit quad-core processor)
  • ($ 35) Audio cables and a waterproof box installed near the shoreline
  • ($ 75) Power over Ethernet -- cable+splitters from the box to a 24-hour UPS and Internet router

60% of cost is hydrophones! (>75% for least expensive research-grade)

DIY: How to build your own hydrophone node!

New node software

Can we stream using open-source software?

  • Linux (Raspian), Docker (container), Dataplicity (remote access), and soon Resin.io (for deployment)
  • ffmpeg -- encodes audio data (compressed stream and/or lossless archive) in two formats:

    How do we archive data in the cloud?

    • s3fs -- uploads data files and "manifest" to Amazon’s Simple Storage Services (S3)
    • FLAC filenames contain UTC timestamp

    Orcanode Github repository

Orcasound app

Can we make it really easy to listen (with a web app)?

Heroku-hosted web site implemented in Phoenix/Elixir: live.orcasound.net

Orcasite Github repository

Orcasound app

Yes, it's getting much easier to listen on any device!

50% of beta and live users report "It just works!"

Failure or frustrated reports down from 32% in beta, to <25% in public live site

Show of hands: did you hear Haro Strait? (Yes? No?)

Global reach

Users per country during 4 days after launch of live.orcasound.net

Local reach

User geography in WA over 4 days post-launch

How can we best set the stage for machine learning?

Orcasound app features under development

  • Beta-test "I hear something" button
  • Build administrative interface to human detections
  • Develop spectrogram visualization of live HLS stream
  • Build training data set of killer whale signals for machine learning
    1. Organize expert's catalog of signals (~40 types of "pulsed calls")
    2. Clip isolated, clear calls from continuous recordings of wild killer whales

These will be foci for the Nov 17 "Hack to Give Thanks" event in Seattle...

Live stream precedent: OrcaLab's OrcaLive

Real-time inspiration: ~5 decades of live-stream pioneering by Paul Spong and Helena Symonds.

Live-stream

Annotation precedent: Orchive

Live stream precedent: Mars Mooring via YouTube

Live stream precedent: Listen to the Deep

Live stream precedent: Cornell's real-time bird call classifier

Human classifier precedent: Zooniverse classify tool for Bat Detective

Orcsound's road map to detection, classification, & annotation

Questions?

← →

/