The AI program for recognizing bird vocalizations, Merlin, is available on your phone. It's really changed how birding is done. It will detect and identify the species of birds in real time as you are walking around, often before you yourself recognize them (if you are even able to).
I've found it to be quite accurate here in upstate NY, with a few recognizable exceptions. In particular, sometimes it seems to think it's hearing a blue-headed vireo when a red-eyed vireo is singing. Also, it sometimes misreports yellow-rumped warblers.
How about an amplifier connected directly into the Merlin bird app’s always listening mode. You don’t even have to review the files. You’ll just get a list of all the birds it has picked up. Birdnet is probably the same API Merlin uses.
I'd be interested to hear more about this. I came close to working on a similar setup but the primary limitations were power and data connections. We figured we'd either need a manual rotation of batteries and storage, or maybe buried PoE to handle both. We never got past the initial concept
Our slow-but-steady concept is down the batteries and storage fork in the road you lay out. Some areas are too remote for infrastructure, so we just revisit and change batteries and cards periodically.
We use dedicated trailcam-like recorders that drop `.wav` files to a 256GB SD card for a few months on four "C" size batteries. Several sites are very remote (islands separated from mainland by several miles), and recordings are scheduled for only 6 hours per day according to the study's focus.
It's using a cheap (0.40c) sound card, but personally I choose to have high quality mics (primo em272 mic capsules, used in high quality parabolic systems.
However, there's nothing to stop you using < 10.00 euro microphones.
sbts-aru writes the files to an SD card so you don't need a whole running windows for this.
Also, it passes all the audio through real time jackd, so while it's writing files you can also have an addition consumer running a real time matchin algorithm on it. If you use a Pi 4 or 5 you would be able to also do that at the same time on the same Pi as is recording.
PoE works better (cheaper, higher bandwidth, differential signalling for higher EMI resistance, readily shared, etc.). Plastic covers are ~free (splash out and deploy duct tape if necessary). MEMS mic -> I2C -> cheap MCU -> ethernet. Theory: Capture digitally, and send the data - forget XLR and other expensive analog 'professional' fancabulation[0]. [0] https://www.youtube.com/watch?v=Ac7G7xOG2Ag
POE very easily tends to inject noise into microphone inputs. If you run a Raspberry Pi for example with POE you get that. This can be fixed by using a power conditioning hat and a POE splitter but now the costs are really increasing.
Well.. the project wasn’t intended to be an acoustic recording device for biologists. It was intended to be a microphone addition to a security project I have that would dynamically merge audio with video. Something you can’t do with microcontrollers.
But along the way, I added a GPS to keep the time accurate and then with the sub-microsecond time accuracy I thought of sound localization. So I developed it further into a sound localizing recorder with in principle enough time accuracy to localize bats that are close even.
With the power of a pi and sub-microsecond clock timing, coupled with a linkage into jackd that can support multiple consumers, you can both record and still have enough CPU to on principle perform real time bird recognition and localization. That is not going to be possible on a microcontroller.
With this project I have been able to localize explosions to less than 20m accuracy from microphones located more than 3km away. Again not something possible on microcontrollers.
The setup of the author is at his own home. Which means that in principle he’s not constrained by power. With that’s setup by adding icecast2 and darkice. Whilst still recording and not needing a big host host computer. Likely even better with go2rtc. Next on my list to try.
Why do you need to merge these items at the point of collection? Why not just stream the sensor data off with temporal correlation data and assemble them elsewhere?
Either way, galvanically isolating 5W isn't too terribly hard either.
Streaming it away and then merge is exactly how it's intended to with jackd streams. You create a dummy slave off the first one, then you have clients off that.
The point being that if I had take a potentially easy way of just recording directly from the device to write the files, then I would not easily at the same time be able to perform real time analysis of the stream. So the way I've done it you can indeed do this.
Note, with manual examination of the sound files with raven lite, I'm able to reliably determine the start time of a sound to less than a millisecond of error (Jitter in the pipeline, otherwise it would be a lot more accurate).
Making it entirely possibly to locate the position of bat calls in real time to quite a good degree of accuracy if you wanted.
The chip you point to suggests a much simpler approach. Is it really that simple? There are a great many more bits on the board I point to. Why would they take a hard way if one simple chip can do it ?
The board you linked isn't a galvanic isolator, it's just a capacitor bank covering a bunch of different frequency ranges.
The ADUM isolator I linked as well as flyback converters use transformers to provide a high-voltage isolation barrier between the source and device being supplied.
Transformers are generally larger, more expensive, and less efficient than non-isolated DC circuits.
The POE supply itself is isolated (with a transformer) and doesn't need to be noisy necessarily - but if you want even more quiet, another level of isolation can't hurt.
I suspect that though the author describes running a USB cable to the house, he may be preparing it for remote operation.
(If you can count on it being next to a house where you can run a USB cable, you could also just power it off USB and skip the battery, right? So the battery must be designed in for a reason.)
The mic is a condenser mic, so a pre-amp is needed, XLR or not.
Cheap mics tend to be single-ended things without the balancing transformer or amplifier, so that would all be extra cost. XLR connectors and cables are more expensive. You'd want to go into some audio interface with XLR jacks which is also more expensive.
'remote' also means you can set up networks of them to cover your area. Just dump a bucket every n meters, and spend a few hours driving around every few weeks to harvest the data & swap out batteries.
Are there microphones that would let you "listen" to bats? We have bats fly around our house at dusk, it would be interesting to know what they "say", especially when two of them come close (unclear whether that's intentional or not).
Reading wikipedia "Bats emit calls from about 12 kHz to 160 kHz, but the upper frequencies in this range are rapidly absorbed in air.", so I guess the Primo wouldn't be suitable for the higher bat freqs.
If you want to pay more you can also get Pettersen ultrasonic microphones. These have a USB microphone and will also work with my sbts-aru recorder. Which means you could in principle also geolocate where the bats are.
The AI program for recognizing bird vocalizations, Merlin, is available on your phone. It's really changed how birding is done. It will detect and identify the species of birds in real time as you are walking around, often before you yourself recognize them (if you are even able to).
Merlin is fantastic!
But it it's not 100% accurate and currently focuses on birds from Canada, the US, and Europe.
I've found it to be quite accurate here in upstate NY, with a few recognizable exceptions. In particular, sometimes it seems to think it's hearing a blue-headed vireo when a red-eyed vireo is singing. Also, it sometimes misreports yellow-rumped warblers.
How about an amplifier connected directly into the Merlin bird app’s always listening mode. You don’t even have to review the files. You’ll just get a list of all the birds it has picked up. Birdnet is probably the same API Merlin uses.
I’d assume so, they’re both from Cornell’s Lab of Ornithology. The Merlin app is incredible; seeing the calls identified in realtime is magic.
I collaborate on a project like this, but using a distributed network of a dozen or so dumb recorders within 50 miles of my house.
I'd be interested to hear more about this. I came close to working on a similar setup but the primary limitations were power and data connections. We figured we'd either need a manual rotation of batteries and storage, or maybe buried PoE to handle both. We never got past the initial concept
Our slow-but-steady concept is down the batteries and storage fork in the road you lay out. Some areas are too remote for infrastructure, so we just revisit and change batteries and cards periodically.
do you use the same hardware? I'm interested in helping a local ornithologist setting one up.
We use dedicated trailcam-like recorders that drop `.wav` files to a 256GB SD card for a few months on four "C" size batteries. Several sites are very remote (islands separated from mainland by several miles), and recordings are scheduled for only 6 hours per day according to the study's focus.
ty
Can you elaborate on your setup?
Why not keep all the electronics and sound card in the house and run a long XLR cable to the microphone?
Battery and sound card and preamplifier outdoors to then run a USB cable out to it seems like a bit of a faff.
I developed a recorder project last year that make's this easier and you could deploy it on a Raspberry Pi zero and use a wifi link.
https://github.com/hcfman/sbts-aru
It's using a cheap (0.40c) sound card, but personally I choose to have high quality mics (primo em272 mic capsules, used in high quality parabolic systems.
However, there's nothing to stop you using < 10.00 euro microphones.
sbts-aru writes the files to an SD card so you don't need a whole running windows for this.
Also, it passes all the audio through real time jackd, so while it's writing files you can also have an addition consumer running a real time matchin algorithm on it. If you use a Pi 4 or 5 you would be able to also do that at the same time on the same Pi as is recording.
Checkout how good the primo mics are here for example:
https://acousticnature.com/journal/how-to-make-diy-microphon...
PoE works better (cheaper, higher bandwidth, differential signalling for higher EMI resistance, readily shared, etc.). Plastic covers are ~free (splash out and deploy duct tape if necessary). MEMS mic -> I2C -> cheap MCU -> ethernet. Theory: Capture digitally, and send the data - forget XLR and other expensive analog 'professional' fancabulation[0]. [0] https://www.youtube.com/watch?v=Ac7G7xOG2Ag
POE very easily tends to inject noise into microphone inputs. If you run a Raspberry Pi for example with POE you get that. This can be fixed by using a power conditioning hat and a POE splitter but now the costs are really increasing.
But it works.
Galvanically isolate the ~100mA of supply current you need - the Ethernet interface is already isolated.
I’m pretty sure that Raspberry pi’s draw more than that. But if you have a good way to fix this I’d like to know.
The pi supply hat also has that probably and no one has fixed it in a couple of years.
Yeah: don't use a Raspberry Pi or any similar Linux SOC to translate sensor data to TCP/IP.
GP said "cheap MCU" - which I agree with. Ambiq MCUs use 1mA while active, but most ARM M0+ cores (RP2040, STM32) are going to be sub-50mA generally.
Spin a $2 4L PCB with all the POE circuitry (and receptacle), MEMS microphone IC, and MCU onboard.
This whole thing should cost <$25 total.
Well.. the project wasn’t intended to be an acoustic recording device for biologists. It was intended to be a microphone addition to a security project I have that would dynamically merge audio with video. Something you can’t do with microcontrollers.
But along the way, I added a GPS to keep the time accurate and then with the sub-microsecond time accuracy I thought of sound localization. So I developed it further into a sound localizing recorder with in principle enough time accuracy to localize bats that are close even.
With the power of a pi and sub-microsecond clock timing, coupled with a linkage into jackd that can support multiple consumers, you can both record and still have enough CPU to on principle perform real time bird recognition and localization. That is not going to be possible on a microcontroller.
With this project I have been able to localize explosions to less than 20m accuracy from microphones located more than 3km away. Again not something possible on microcontrollers.
The setup of the author is at his own home. Which means that in principle he’s not constrained by power. With that’s setup by adding icecast2 and darkice. Whilst still recording and not needing a big host host computer. Likely even better with go2rtc. Next on my list to try.
Why do you need to merge these items at the point of collection? Why not just stream the sensor data off with temporal correlation data and assemble them elsewhere?
Either way, galvanically isolating 5W isn't too terribly hard either.
Streaming it away and then merge is exactly how it's intended to with jackd streams. You create a dummy slave off the first one, then you have clients off that.
The point being that if I had take a potentially easy way of just recording directly from the device to write the files, then I would not easily at the same time be able to perform real time analysis of the stream. So the way I've done it you can indeed do this.
Note, with manual examination of the sound files with raven lite, I'm able to reliably determine the start time of a sound to less than a millisecond of error (Jitter in the pipeline, otherwise it would be a lot more accurate).
Making it entirely possibly to locate the position of bat calls in real time to quite a good degree of accuracy if you wanted.
Because you can. Why split the architecture across multiple machines when it works on one.
Do you have any recommended links to that galvanic isolation I could look at ? I’m encouraged by your view on this.
I don't really want more Linux boxes floating around if I can help it.
https://www.analog.com/en/products/adum6000.html
Things like this are a fancy way to go about it - otherwise very-filtered flyback will probably get you where you need to be.
I've tested this device and that works. Nice clean power on the microphones whilst running over POE.
https://thepihut.com/products/power-filter-board-for-raspber...
The chip you point to suggests a much simpler approach. Is it really that simple? There are a great many more bits on the board I point to. Why would they take a hard way if one simple chip can do it ?
The board you linked isn't a galvanic isolator, it's just a capacitor bank covering a bunch of different frequency ranges.
The ADUM isolator I linked as well as flyback converters use transformers to provide a high-voltage isolation barrier between the source and device being supplied.
Transformers are generally larger, more expensive, and less efficient than non-isolated DC circuits.
The POE supply itself is isolated (with a transformer) and doesn't need to be noisy necessarily - but if you want even more quiet, another level of isolation can't hurt.
Interesting. Do you know of any built products that use the approach you are suggesting ?
Olimex ESP32 POE ISO with a PDM or I2S microphone breakout board is an easy version without having to create a custom board.
I think you mean I2S (or PDM)? I2C is not common on MEMS microphones, as the bandwidth is generally insufficient for audio signals.
Right, I don't usually touch audio or video applications.
I suspect that though the author describes running a USB cable to the house, he may be preparing it for remote operation.
(If you can count on it being next to a house where you can run a USB cable, you could also just power it off USB and skip the battery, right? So the battery must be designed in for a reason.)
The mic is a condenser mic, so a pre-amp is needed, XLR or not.
Cheap mics tend to be single-ended things without the balancing transformer or amplifier, so that would all be extra cost. XLR connectors and cables are more expensive. You'd want to go into some audio interface with XLR jacks which is also more expensive.
'remote' also means you can set up networks of them to cover your area. Just dump a bucket every n meters, and spend a few hours driving around every few weeks to harvest the data & swap out batteries.
Or even better do it an isolated area to filter as much noise as possible
Are there microphones that would let you "listen" to bats? We have bats fly around our house at dusk, it would be interesting to know what they "say", especially when two of them come close (unclear whether that's intentional or not).
I've used Primo EM258 mics with an XLR adapter, they can pick up ultrasound, although the datasheet doesn't really mention this.
See - https://zachpoff.com/resources/cheap-microphones-for-ultraso...
Reading wikipedia "Bats emit calls from about 12 kHz to 160 kHz, but the upper frequencies in this range are rapidly absorbed in air.", so I guess the Primo wouldn't be suitable for the higher bat freqs.
There are!
https://www.omenie.com/pipistrelle.html
Very nice project the above.
If you want to pay more you can also get Pettersen ultrasonic microphones. These have a USB microphone and will also work with my sbts-aru recorder. Which means you could in principle also geolocate where the bats are.
See also https://app.birdweather.com/ which uses a community of stations using AI to match vocalizations with bird species. It was pretty easy to setup a Pi with a microphone using https://github.com/mcguirepr89/BirdNET-Pi to start identifying birds outside my place.
That plastic dish (parabolic reflector) is not cheap.
They link to Amazon and it's whopping $8.99.
It would be, if it wasn't just a "plastic bird-feeder cover" as the article suggests