Dear Robin,

So what is your workflow regarding the utilisation of birdnet to analyse sound bird?

All the best,

Hi Daniella,

I don't know specifically about AudioMoth, but people have had very good results reducing or eliminating wind noice on things like GoPros or iPhones, etc. using simple Microphone Windscreen foam.  You can buy ones for headsets and the like pretty cheap (10 for $1 for the little ones).  You can just cut them up and either tape them in place on the inside of the case or rubber band them on the outside.  

Also, if the sounds you are trying to extract are still audible, but hard to pull out of the noise, you might also be able to post-process the wind noise out.  Wind tends to be heavy in low frequencies, so depending on what you are looking for you might be able to just filter the lower frequencies out, or use an open source tool like Audacity.  But if your signal is buried deep within the noise, these tools might also corrupt your target signal.

Hope this helps.

Drue

I second what Drue said on both fronts. Using a windscreen for any microphone is really helpful in reducing wind interference. For the Audiomoth's size, you could probably use a lavalier mic windscreen of some sort inside your case. I post-process wind interference out sometimes using bandpass filters in RavenPro (cutting out low frequencies <1kHz usually gets most of it at the site I work at) but this will depend on the frequencies of your target sounds. 

furry "dead cat" covers and blimps work best for cutting down wind noise, and a very good spectral noise repair tool, though it's paid, iZotope RX8. There is a de-wind module, but you can also teach it what to repair. I believe Davinci Resolve and Adobe Audition might also have wind reduction tools, but my go-to is iZotope RX8. 

Stuart, that is really cool! Are these clips translated into audible frequency range?

Ah, perfect. I was wondering why I never hear those calls!

That's really nice work, thanks so much for sharing!

this guys has been around a long and has some neat stuff http://songsofinsects.com/about

but sure looks like a wide-open area of research ...

Tom Dally at Leeds University UK is doing work in this area. I dont think he has anything published yet but his preliminary results are interesting (saw them at a workshop).

https://biologicalsciences.leeds.ac.uk/school-of-biology/pgr/753/thomas-dally 

Tom

Thanks everybody for joining today's session! I've attached the article here that describes Pattern Matching in greater detail to respond to some of the questions from today's session. 
Also, here is a link to our Privacy Policy and our Terms of Service for the individual who had questions about GDPR. Feel free to contact us at [email protected] with any other questions about the platform or Pattern Matching in general! 

Zephyr and Marconi, thanks for the great talk and your hard work that made this platform available to the public!! I'm experimenting with the pattern matching function to create some training data but have issues trying to get the "Jobs'' to run. I only have 18 files for each of these tests, but only 1 processed to 11.1% while two others stay at 0%. The internet connection was briefly interrupted when I created the first job and was fine later on. I can only Hide, not delete or re-run these jobs, so I'm not sure what to do. Any suggestions would be greatly appreciated. Thanks so much!

I’ll stick some thoughts here that other folk can maybe comment on/add to. 

So there are a few (obvious) differences between terrestrial and marine acoustics. 

• Devices are immersed in a highly corrosive and conductive liquid (sea water) with strong underwater currents, fishermen, storms etc. 
• Boat’s are usually required to deploy devices - that can mean it is much more expensive to make recordings. 
• Ultra high frequency species and noise sources (e.g. harbour porpoise @130kHz and broadband delphinid clicks @ >200kHz and echo sounders @ 200kHz) are relatively common. 

For PAM devices that means a few things: 

• Longevity is crucial because it can greatly decrease the cost per hour of recordings (less boat time is required). There are a few strategies for maximising this;

1. Devices such as FPODs and SoundTraps run onboard detectors. The SoundTrap in particular can run a click detector at high frequency (up to 576khz sample rate) whilst also contiously recording at lower sample rates.  This means that it detects high frequency delphinid and porpoise clicks (and saves the waveform for further analysis) and at the same time records all lower frequency sounds where almost all of the complex tonal vocalisations of dolphins, baleen whales and anthropogenic sounds occur; these types of sounds also happen to be more difficult to accurately automatically detect (especially on low powered hardware). This strategy results in an order of magnitude reduction in data without incurring a large cost in data degradation. 
2. Run on board compression algorithms - the SoundTrap uses an open source X3 compression algorithm that often results in four times lossless data compression. 
3. Have the space for a lot of batteries and/or the option of an external battery pack. 

• Sensitivity is important because you want to maximise your monitored area (again making your survey as efficient as possible). That means choosing a hydrophone, amplifier and DAQ with a low noise floor and appropriate clip level. 
• Devices must be robust and waterproof. They should be able to survive biofouling, getting knocked on the seabed and ending up in a fishing net etc. The two common housing designs are to have a completely sealed unit with an underwater connector (often SubConn) for downloading data or having a unit which can be opened to remove memory cards and batteries. Both approaches have their advantages and failure points. For example, devices which are sealed tend to be more simple to set up but the connector is often a failure point. Devices which can be opened are serviceable at sea but they can flood and are air filled which makes them a bigger acoustic target for echolocators (and so can mess up results). 

There are also a few ‘nice to have’ options. 

• Two sample synchronised hydrophones allowing a bearing to be calculated. This can be useful for a whole host of stuff - for example using bearing tracks to figure out how many animals might be present, improving density estimation methods and allowing an extra data dimension for classification algorithms. It also means the devices are useful for a host of other applications - e.g. towed hydrophone surveys. 
• Depth, temperature, tilt and light sensors are inexpensive low power and produce only a few gigabytes (at the very most) of data over a survey but can provide a wealth of oceanographic data that is also helpful in acoustic monitoring. 
• If running detectors, then recording noise is critical because it allows us to model changes in the probability of detecting animals which is often important in density estimation. 
• Accurate clocks are useful. The  SoundTrap clock drifts around 2s per day. Although manageable, this is often a pain and consumes valuable analysis time re-syncing devices together. 
• The ability to daisy chain devices so they can be sample synchronised and used in acoustic arrays is very handy. For example SoundTraps have been used to make large aperture hydrophone arrays to localise Kogia. DOI: 10.1016/j.dsr.2020.103233 (if achievable this may supersede point 1. )

Ok, that’s everything I can think of for now. I would be keen to hear other opinions on this, additions, disagreements etc.  Jamie