Hands On With Boondock Echo

Perhaps no words fill me with more dread than, “I hear there’s something going around.” In my experience, you hear this when some nasty bug has worked its way into the community and people start getting whatever it is. I’m always on my guard when I hear about something like this, especially when it’s something really unpleasant like norovirus. Forewarned is forearmed, after all.

Since I work from home and rarely get out, one of the principal ways I keep apprised of what’s going on with public health in my community is by listening to my scanner radio. I have the local fire rescue frequencies programmed in, and if “there’s something going around,” I usually find out about it there first; after a half-dozen or so calls for people complaining of nausea and vomiting, you get the idea it’s best to hunker down for a while.

I manage to stay reasonably well-informed in this way, but it’s not like I can listen to my scanner every minute of the day. That’s why I was really excited when my friend Mark Hughes started a project he called Boondock Echo, which aims to change the two-way radio communications user experience by enabling internet-backed recording and playback. It sounded like the perfect system for me — something that would let my scanner work for me, instead of the other way around. And so when Mark asked me to participate in the beta test, I jumped at the chance.

Hardware Design

If Boondock Echo sounds familiar, it’s probably because we’ve written about it a few times already. The project was entered in the 2022 Hackaday Prize, and actually ended up winning the 4th place award that year. There’s also a Hackaday.io project, and as a personal point of pride, Mark and Kaushlesh Chandel, now the firmware architect on the project, started their collaboration after meeting at a Hack Chat back in 2021. We get results!

One of these things is not like the other. There are different hardware codecs under the RF shield on what appear to be identical ESP32-A1S modules.

The Boondock Echo beta test units I have are based on an AI-Thinker ESP32 Audio Kit, a commercially available dev board that’s specialized for digital audio applications. It wasn’t the team’s first choice for a platform, not by a long shot, and the path to it is an object lesson in the often harsh realities of product design. The original design called for the same audio module as the full AI-Thinker dev board, the ESP32-A1S. They designed a nice PCB around it with all the interface electronics needed to support a Baofeng handy talkie along with a good-quality speaker, and ordered up a bunch of the ESP modules.

However, they found that despite identical part numbers, FCC ID numbers, and external arrangements, not all the modules they had ordered for the first run worked as planned. It was only when they took the RF shielding off the modules that they discovered the reason: significant differences in board layout and components used, particularly in the area of the ES8388 audio codec.

From the sound of it, sorting all this out was a huge time-sink for the team. With no way to tell from the outside whether a module would work in their design, they eventually had to cut their losses and go with the pre-built Audio Kit board with a small, custom-built “sidekick” board that connects to the Baofeng. It’s a sensible approach, especially given the circumstances, and not only does it allow them to actually ship working units, but it also gives them a chance to move to another, more capable platform later.

They Don’t Call It “Easyware”

From the outset, let me make it clear that I suspect I’ve been anything but the typical beta tester, at least from the team’s point of view. Boondock Echo was designed primarily to retransmit the audio it receives; it’s right in the name, after all. Mark’s original vision for Boondock Echo was based on his experience with California wildfires and the spotty radio communications that often interrupted the timely flow of news and evacuation orders But my immediate interest in Boondock Echo is for recording public safety radio calls from my scanner. So, rather than just plugging into the Baofeng handy talkie they sent along with my test unit, I had to go rogue and try my own thing. Sometimes you just have to break things.

And break things I did. Try as we might, the first Boondock Echo I got just wouldn’t work. We tried everything — punching holes in my firewall, disabling my Pi-Hole ad blocker, and even setting up a hotspot on my phone to bypass my home WiFi. That last attempt actually worked, meaning that the was something specific about my network that was making it impossible for the device to register with the Boondock Echo server. This resulted in some very long conference calls with Mark, KC, and hardware architect Jesse Robinson, and kudos to them for trying everything to get me online. But they simply had to go back to the drawing board.

Things went much better on the second attempt. They sent me a new unit with updated firmware running on essentially the same hardware, with an updated and much-improved enclosure. I was able to get this device online with very little effort; all that was required was editing a config file on the included SD card with my WiFi credentials and rebooting. However, audio from my scanner — a Bearcat BCD996P2 — was not being recorded. Again, the team was very professional about it, getting on a conference call with me and later continuing the discussion over a Discord chat. We eventually figured out that the line-out jack on my scanner just wasn’t putting out as much signal as the Baofeng external speaker jack does. KC went off and changed the firmware to increase the dynamic range of the audio input, and between that change and a little bit of adjustment of the attenuator on the side-kick board, I was finally recording audio.

My Experience

At least in my case, once the Boondock Echo is online, there’s not much need to interact with it. That doesn’t mean you can’t, of course — there are buttons on the front panel that let you control the volume of the internal speaker, scroll back and forth between recorded sound clips, and even a push-to-talk button that will record a clip through the front panel mic for transmission on the Baofeng, if you’re suitably licensed to do so. There are also some LEDs that tell you the current status of the device. All these are thoughtful features, but in practice, I found that all the real action is in the Boondock Echo web interface.

To be clear, the Boondock Echo application isn’t running on the ESP32 built into the device. Rather, it’s hosted on AWS, for security and scalability in the future. The ESP32 in the device takes care of all the local functions, like digitizing the audio, running the local interface, and playing back audio clips locally. I find the design of the web app pretty good; it’s functional without a lot of flash or fluff, but still good-looking and intuitive. The main page has a list of audio clips that have been recorded by your device — or devices; the idea of a “dock pack” of multiple Boondock Echo devices is supported, for capturing audio from multiple sources. Each entry has an embedded player to let you listen to the clip on your PC or phone, along with buttons that send the clip back to the Boondock Echo device for playback over the internal speaker, or for retransmission over the attached transceiver. Again, a proper license is required to transmit.

One of the neatest features of Boondock Echo, and the one that I was most interested in testing, is online transcription of recorded clips. Clips are sent to OpenAI’s Whisper API for transcription, and all things considered, the results are pretty good. There are obviously limits, of course; it doesn’t seem like Whisper’s training model includes such catchy phrases as “patient contact green,” which is used around here when an arriving unit encounters a patient who isn’t in any immediate danger. Transcription of specialist speech like this is spotty, with results like “face your contact green,” or “Trojan 41” for “Engine 41.” I’ve also noticed that when in doubt, Whisper seems to skew toward the very polite, and even a bit amorous; there are plenty of “Thank very much” and “Have a great day” transcripts, which I’m reasonably sure isn’t what was said, and the occasional “I love you” sprinkled in, especially on channels where users tend to talk way too loud.

That brings up a good point, and one that there’s probably not much the Boondock Echo team can do much about right now. The quality of audio in public service radio can be all over the place, ranging from barely audible to over-driven and distorted. Certain dispatchers are soft-spoken, others louder and more confident on the air, while firefighters and paramedics often get into situations where they can’t help but get a bit excited. There are also what I call “incidental radio operators,” such as the nurses who talk to paramedic units in the field and take reports on the radio. These people are medical professionals without much radio training, and they all tend to “swallow the mic,” resulting in severe distortion.

The team hopes to implement some DSP algorithms to clean up audio before it goes off to Whisper, but that may have to wait until they make some hardware changes to the Boondock Echo. That’ll be a huge benefit to one of its most useful features: keyword monitoring. Users can set up monitors that scan transcripts for specific keywords and then perform an action. In my case, I have monitors set up for the names of roads my family and friends live on; if there’s an ambulance or fire call for my parents’ address, I want to know about it instantly. At this time, notification when a keyword matches is limited to an email, but future actions will include SMS texts and even IFTTT, so you can do pretty much anything you can imagine. It all depends on the accuracy of the transcription, though, since the match with the keyword needs to be exact. It might be a good idea to implement the Soundex algorithm here, or at least make it an option.

Hiccups aside, I’ve been very pleased with how my Boondock Echo has performed, especially considering that it wasn’t built to support my specific use cases. I’m still very early in the process of integrating it into my workflow, but I can see a lot of potential uses for it. I can imagine using this as the basis of a local information network, for people who don’t want to sit in front of a scanner all day listening to all the routine stuff, but really want to know when “the big one” hits. I’m also keen to try setting up a “store and forward” simplex repeater for ham radio experiments; a Boondock Echo operating in offline mode would make this simple to set up.

Hats off to the whole Boondock Echo team for going from idea to working project in such a short time, and thanks for inviting me along for the ride. I’m looking forward to seeing where it goes next.

22 thoughts on “Hands On With Boondock Echo

      1. Now THIS sounds actually useful. A pity it didn’t win first place, but then that seems to have been the nature of things up to this year.

        I honestly can’t remember if I’ve criticized this entry in the 2022 Prize — if I did (likely) — I apologize. It’s kind of easy to pour on the salt with that… especially when it gets swept under the rug without so much as a howdy-do. (Did you /really/ think I’d not notice when my comment went missing? I’ve earned my time to gloat, thank you just the same.)

        The two things that immediately stand out are (a) usability issues and (b) cloud reliance. The fact that simple things like misconfigured audio or a slightly unfriendly WiFi network sounds problematic to me. That said, judging by the author’s experience regarding firmware updates, at least the network issues are a known problem undergoing active development to fix. I hope something to handle audio levels is likewise under consideration.

        Remember, you can’t rely on users, even techy types, to submit bug reports or email the dev team or what-have-you, every time something trips them up. They’re far more likely just to drop and leave. I’m guilty of that myself…

        The use of Amazon cloud services, though, that really concerns me. Up to a point it works, at least in most urban areas — but not all of them, and not in a lot of rural areas. I live in central North Carolina. There are friends of my father who live in Raleigh, and their internet choices, right now today, are (a) 56k dialup or (b) physically going to the nearest coffeeshop with free WiFi. That is not a joke, that is reality.

        Other friends of mine can’t afford home Internet on a reliable basis, or at all in some cases — Raleigh’s quite a drive away, out here in Nowheresville County, USA — and this is just NC. I don’t even want to think about states like Utah or Kansas or Wyoming or Nevada where it can easily be a couple hours’ drive simply to get to the next gas station flush toilet, let alone something properly resembling modern civilization.

        This sort of thing needs to be able to operate locally, feeding off broadcast RF transmissions and processing data on-site. I could easily see something like COVID bringing it down simply due to bandwidth cap issues, if you’re relying on cloud services… never mind if something worse were to go down, heaven forbid.

        Honestly, I never thought I’d say it as long as I lived, but this is a legit use for a RasPi. You need something with microcontroller-style IO, but proper computer processing horsepower, and it’s an application where you REALLY don’t want it to power-cycle. A Pi with some sort of battery backup behind it would actually be really useful here. For years I’ve been saying the Pi is a solution in search of a problem… ladies, gentlemen, and [etc], here’s the one problem I’ve ever seen that it’s a valid solution for.

        Incidentally, this is a really wholesome project just in general, and I’d like to see Hackaday cover updates from it a lot more in the future.

        1. Thank you very much for your comments! I came in late to the project, from what I’ve been told the pi was unavailable when they developed, and KC is very familiar with the esp32 and the audiokit dev board was available. We are definitely eyeing up the pi for a future generation.

          As for the network and cloud, the original goal was being a ham radio answering machine that would store and playback messages, KC is just really good at iot and started slipping in more and more cloud and AI features. You’d be surprised how good they are and reliant you become on them. Like I leave my dock running all the time and then check it with my phone at a friend’s house or how I prefer to read the transcription for a quick synopsis of the message. But I see your point about network connectivity.

          One thing I’m excited for is the dock packs where you can link docks together. So you can talk on one and have it transmit on a dock somewhere else physically. So you could use the Internet to hop around challenging terrain.

          Feel free to sign up on crowdsupply to follow the journey or join the discord you chat during development.

          1. Thank YOU for such a lovely response!

            This is one place where I have to say I’m going to champion scope creep. I find it almost impossible to feel anything optimistic about the original purpose as described — a sort of ham radio answering machine. But something that warns people that there really IS something in the air or water that might be of concern? I’m immediately behind that 1000%.

            I believe I’m technically on CrowdSupply because of a project that seems to have somewhat stalled out, entirely unrelated to anything Hackaday’s ever covered, that’s kind of interesting in concept but seems to have reached its own internal heat-death as far as age and passion goes. That said, I’ve nothing myself that I could really offer by way of assistance, and I /hate/ the feeling of being a third wheel… and I’ve *ahem* a bit of a learned fear of Discord servers and their ilk, so I’ll not be joining up — but I’ll once again reiterate the hope that Hackaday will choose to dip in from time to time and run an update feature on the project once in a while.

        2. Thanks a lot for your response. Great feedback. As we started using it, we did notice most of points you mentioned, so these things are in our priority list, and roadmap.

          1. Audio levels for different radios
          2. Simplicity of use. User plugs it in, and it should work
          3. Offline mode. it keeps recording with no internet, and syncs up when internet resumes
          4. No internet? In future ( Pi 5 ??? ) we would love to build fully offline version

          We have open-sourced the code. Hopefully by the power of community, we will see development speed up.

          This project started as a problem solver. with enough support, we will be able to pay the bills and keep the lights on :). We have had a lot of awesome feedback, and there is a lot more to come.


          1. Sounds like you’ve got a good idea of what you need to address where and when. Not all such projects can claim such clarity — well done!

            I wonder if a Pi3 or Pi4 would be enough for the offline version…? I know the Pis have been pretty severely affected by that chip shortage thing but I’m not sure how much, because I’ve tried to stay away from that particular drama as best I can.

            Regardless, as I’ve already said, I look forward to what this project brings in the future. By all means, keep the lights on and the soldering iron going!

        1. Thanks for the clarification. I’m sure a device made to be used in certain emergency situation can only beneficiate from a totally offline mode (with a local computer to do the heavy work, not a faraway internet server that could be unavailable in an emergency situation (no internet))

  1. Seems a heck of a lot more useful than using my Signalink USB and Audacity!
    I wonder if the keyword feature would be discerning enough to accurately listen for certain call signs so I could listen for my friends without having to keep my ear glued to my radio.

  2. I know these online ‘AI’ services can be very useful and it’s trendy to integrate it in to projects, but I’m a little disturbed by how readily people are to send off images (and in this case audio) of random strangers off to these private corporations for processing. A week ago there was a Halloween project where someone was sending off pictures of (presumably) kids coming to their door to be online-face-recognition’ed. No warnings, to chance to opt-out or avoid it.

    Yes, this is all sourced from ‘public’ spaces, but have people really become that apathetic towards the privacy of others?

  3. Home Assistant is now doing about 80-90% of this already. 2023 has been their Year of the Voice, and the results are pretty impressive. Their voice recognition system is also based on Whisper.

    And it works. I already have local voice recognition running in my house, where I’m hosting the entire voice stack locally. The software components I’m running are OpenWakeWord (listening for “Hey Jarvis”), Whisper (speech-to-text), and Piper (text-to-speech). In my living room I have a tiny $13 ESP32 Atom with integrated microphone and speaker, which I flashed from their site with no customization other than pointing it at my WiFi. For the most part it works as long as the audio environment is quiet, I speak clearly, and I stick to the scripted phrases.

    Since the team has kindly open sourced their code, it seems like a “Small Matter of Porting” to divert the Boondock Echo’s audio stream to a local instance of Whisper. But it’s going to be lacking their all important server work that identifies and stores the audio.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.