Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Listening for illegal logging chainsaws using TensorFlow

blog.google

Post reply on HN
Print view View original

121–128 of 128 posts

Re: Listening for illegal logging chainsaws using TensorFlow

#121

Post by smarx007 »

smarx007 wrote: post #120
stefanRfcx wrote: post #35

Earlier quoted context omitted.

Thanks! We are a relatively small engineering team and we were mainly focused on improving our system. This took a lot of effort and time, we just haven't had a chance yet to think about open sourcing or blogging. Our days were focused on coding. If there's interest in blog posts, I could write up some of our technology. Just let me know whether there's an interest and what you'd like to know.

Just signed up for your newsletter! I understand you don't want to open up the audio fingerprinting details, but everything else would be very interesting. I am interested in how do you get an old Android phone to run some your APKs headless, how do you update them OTA, how do you power them with solar (I guess you have an external power bank) — some basic questions if I want to put an old Android phone to my DIY pro…

I don't use Android but just did some googling and only have these ideas: roll your APK updates over a custom F-Droid repo, use SSHDroid to log into the headless machine, and/or use one of the DEX patching libraries. How far off are my thoughts?

Re: Listening for illegal logging chainsaws using TensorFlow

#122

Post by LifeLiverTransp »

John_KZ wrote: post #119
LifeLiverTransp wrote: post #112

Earlier quoted context omitted.

The sole voice of reason- thank you. This money should be invested into fusion or solar research - if energy is cheap- somebody will find a vay to convert it into fast vat grown rainforrest- and voila-> Unemployed drunks protesting before the automatic growthlab and sawmill. Its very hard to prevent behaviour thats economical. Its way easier to subvert the economics behind that. If you want to save nature- compress a…

>Unemployed drunks protesting before the automatic growthlab and sawmill. Or politicized unemployed people chopping your head off because you took their jobs and gave nothing back. It goes both ways. I don't think you understook what I meant. By giving them an alternative source of income, I meant the workers, not the corporations.

Maybee they can get jobs as click-AI- basically providing non-savant intelligence to AI-Projects.

But my point still stands, if growning local and fast in vats- rainforrest wood would be no more economical viable.

No pressure on any fusion researchers- but you could really save the planet right about now.

https://www.straightdope.com/columns/read/3018/whatever-happ...

Re: Listening for illegal logging chainsaws using TensorFlow

#124

Post by tony_cannistra »

aerique wrote: post #113
ada1981 wrote: post #45

I love these guys! I helped advise them on their crowdfunding campaign a few years ago. One of my favorite projects. Hey Topher! Are you guys cross checking this with daily satellite imagery? I met with a guy who's company gets a full HD shot of the earth every 24 hours and they'd be happy to give you guys access to the visual data (they also train algorithms on it to detect all sorts of activity). Anyhow, I'll shoot…

Do the shots include oceans and seas as well and would ships be visible on those shots? I'm asking because Sea Shepherd could not continue an anti-whaling campaign because they were technologically overmatched: http://www.themercury.com.au/news/politics/sea-shepherd-anno...

Planet does not image the ocean in its entirety due to the difficulty in geo-referencing (pre-user imagery processing pipelines use ground control points, usually just shiny objects with a known location, to locate the image on the earth, and this is hard in the ocean), but they do image focal regions for purposes similar to those you mention (vessel identification + tracking). I'd also imagine this choice has something to do with preserving satellite operational lifespan (can turn cameras off when not being used).

https://www.planet.com/products/open-water-monitoring/

Re: Listening for illegal logging chainsaws using TensorFlow

#125

Post by moltensyntax »

If hypothetically the microphones could be fairly densely spaced, I'm imagining a VR game, where the graphics are dark and indistinct, but the soundtrack is real 3D audio from a Brazilian rainforest. You could add fast simulated motion, could be an intense experience.

Re: Listening for illegal logging chainsaws using TensorFlow

#127

Post by colinnordin »

stefanRfcx wrote: post #43
osipov wrote: post #8

Earlier quoted context omitted.

Yes. In the video there is a short snippet where the audio is shown as a Fourier transformed image on the screen and a user is annotating the image of the sound using red boxes. This is a part of the process to train the ML model to recognize chainsaw sounds vs. other sounds.

Thanks for noticing the spectral analysis. We put quite a bit of work into the training system. Besides the base-level Fourier transformed images, we also have a UI for partners who can easily report if an alert was correct or not which also feeds back into the system.

I also work with non-speech audio and I'm curious: Do you use pure DFT:s as inputs to your models or do you use mel-energies or MFCC:s? What kind of models do you use? Since there is not that much variation in the sound of a chainsaw I suppose either a regular fully connected or convolutional neural network?

Love what you are doing and I would love to see a technical blog post about how you work with audio!

Re: Listening for illegal logging chainsaws using TensorFlow

#128

Post by therealdrag0 »

ada1981 wrote: post #45

I love these guys! I helped advise them on their crowdfunding campaign a few years ago. One of my favorite projects. Hey Topher! Are you guys cross checking this with daily satellite imagery? I met with a guy who's company gets a full HD shot of the earth every 24 hours and they'd be happy to give you guys access to the visual data (they also train algorithms on it to detect all sorts of activity). Anyhow, I'll shoot…

NPR's "Planet Money" podcast did an interesting series about satellites which featured this company, "Planet". https://www.npr.org/sections/money/2017/12/01/567267573/plan...
Post reply on HN