FalconSDR

A wideband radio scanner for SDR receivers that records only the transmissions it actually stops on - no hours of empty noise, no retuning clicks. Point it at any band (aviation, marine, PMR, ham repeaters, public services, military UHF…) and get a searchable library of conversations.

It runs in your web browser: on the computer the SDR is plugged into, or as a headless receiver you listen to from any phone or PC on your network.

You can try UI and UX also in virtual ether (simulated transmissions) before downloading: the real interface on a simulated receiver at falconsdr.com.

Listening to and recording some radio services is restricted or illegal in many countries - see Legal.

Features

Scanning

• Wideband: the whole capture bandwidth of your SDR is analysed at once, so several simultaneous conversations on different frequencies are caught and recorded in parallel, not one after another.
• Any number of bands (start, stop, step, modulation) plus a Memory List of named single channels in folders. Ready-made presets for common bands.
• Priority weights for bands and channels, and a short "recency boost" that revisits a frequency right after it was active, to catch the reply.
• Per-band overrides: trigger level, gain, hold time, squelch, noise reduction.
• Blacklist for carriers and interference, one click from the live view.
• AM, NFM, WFM and SSB demodulation; software or hardware gain control.
• Experimental: several receivers at once - the scan plan is split between them, so each band is revisited that many times more often. They can be of different kinds and widths (say an RTL-SDR at 2.4 MHz next to a HackRF at 10 MHz); the plan is shared in proportion to their capture widths.
• Optional antenna range per dongle (SDR device card): one on a 2 m antenna scans only 144-146 MHz, one on a 70 cm antenna only 430-440 MHz. What no range reaches is listed on the Listen page.
• Optional, experimental digital voice: DMR and Yaesu System Fusion (C4FM) through dsd-fme, installed from Settings → Digital voice. Encrypted calls stay muted.

Listening

• Live audio in the browser, with a frequency display, signal level and a live spectrum.
• Skip, skip + blacklist, solo a single band, listen to a single frequency.
• Works on phones, with play/pause/next controls on the lock screen and headset.

Recording

• One file per transmission (WAV or MP3), named by time and frequency, with a configurable hold time to keep call and reply together.
• Squelch, spectral noise reduction, a cap on simultaneous recordings.
• Manual continuous recording and raw IQ capture (SigMF) when you need them.
• Automatic cleanup by age or disk space, keeping favorites.

Recordings library

• Browse by frequency, band or language; search by frequency, channel name, date or transcript text; filter by time and quality.
• Every recording gets a quality score (signal-to-noise), a waveform and click-to-seek playback with auto-advance.
• Favorites, bulk favorite/delete/transcribe, download a whole selection as one ZIP, keyboard shortcuts.
• Flight replay for aviation recordings (118-138 MHz and military UHF 225-400 MHz): one click opens FlightRadar24 or ADS-B Exchange replaying the traffic around you at the moment of the recording - to see who was talking. Set your approximate location once in Settings → Location.
• Statistics: busiest frequencies, activity by hour, weekday and day.
• Optional speech-to-text with Whisper (installed from the app; one model, large-v3, for every language - an English aviation model can be added in the settings, and fixing the language skips its detection). It can run on a different, faster computer than the one scanning; see what it needs.

Server and multiple computers

• --server mode for a small always-on Linux receiver (mini PC, home server, container).
• Other computers connect to it to transcribe recordings with their GPU and/or keep a verified local copy of the library.
• Optional access password; English and Polish interface (more languages are one translation file away).

Hardware

Receiver	How
RTL-SDR (incl. RTL-SDR Blog V3/V4)	built in on Linux, Windows and macOS
HackRF, Airspy	built in on Windows, .deb, .rpm, AppImage and (since 1.4.1) macOS (experimental)
SDRplay, LimeSDR, USRP, …	through SoapySDR, Linux source install (not tested)

An RTL-SDR is opened through librtlsdr; on Linux from source also through SoapySDR - choose under SDR device → Access method. HackRF and Airspy work through the bundled libhackrf / libairspy (experimental: confirmed on a real HackRF One, untested on an Airspy; pick Access method → libhackrf / libairspy); on Linux from source they work through SoapySDR with the matching module, or through the same built-in access once libhackrf0 / libairspy0 are installed.

HackRF captures 2-20 MHz (the default 2.4 MHz is the safest start); an Airspy only the fixed widths its board has (for instance 2.5 and 10 MHz on an Airspy R2), and FalconSDR picks the nearest. A wider capture needs more processor - on a 4-core mini PC about 1, 8 and 19 ms of every 30 ms block go to the spectrum at 2.4, 10 and 20 MHz, plus 2, 9 and 17 ms for each transmission being recorded. The macOS builds carry the HackRF/Airspy libraries since 1.4.1 (untested on a real Mac).

Developed and tested with an RTL-SDR Blog V4 on Linux and Windows 11. A HackRF One has been checked on Linux (10 MHz capture; SoapySDR and libhackrf both scan, libhackrf reported no lost samples) by phutch in issue #9 - thank you. Other receivers and macOS (experimental builds) should work but haven't been tested yet - reports are welcome.

Testing a receiver: stop the scanner and run falconsdr --probe-sdr (from source: python main.py --probe-sdr; add hackrf_native / hackrf / airspy_native / airspy to pick the way it is opened; on Windows the output goes to %USERPROFILE%\.config\FalconSDR\falconsdr.log). It measures the gain steps, the DC spike, how many blocks after a retune still hold the old frequency, and sample losses, and prints one RESULT {...} line - please paste it into an issue together with your receiver's model.

Any antenna works; one cut or tuned for the band you scan works much better.

Transcription needs more than scanning

Scanning and recording are light - a small mini PC runs them fine. The figures below are only for the optional speech-to-text (it can also run on another computer, see Headless receiver). Pick the model in Settings → Whisper models.

Model	RAM (GPU: VRAM)	Word errors, clean / noisy radio
small - smallest usable	1.4 GB (0.8 GB)	30 % / 51 %
medium - best without a GPU	3.2 GB (1.4 GB)	23 % / 44 %
large-v3 - default	3.9 GB (3.3 GB)	21 % / 38 %

How long transcribing takes, as a multiple of the recording's length, for short ~6 s recordings (at 2× it keeps up as long as you record less than about 30 minutes of audio per hour):

Model	Mini PC, 4 threads*	Desktop, 8 threads†	Desktop, 24 threads†	Nvidia GPU‡
small	1.0×	0.5×	0.3×	0.03×
medium	2.9×	1.1×	0.7×	0.06×
large-v3	6.0×	2.1×	1.2×	0.08×

• A CPU alone is enough to transcribe while scanning: it only has to keep up with how much you record, and a queue that builds up in a busy spell empties afterwards. An Nvidia GPU makes it 5-20 times faster, which pays off for transcribing hundreds of existing recordings quickly or a very busy band. AMD/Intel GPUs are not used.
• Scanning plus transcription: at least 4 CPU threads; transcription uses the ones beyond 4 (1, 4 and 20 in the columns above; at least one), so 8 is comfortable.
• tiny and base are not usable (45-90 % errors). On a CPU turbo is no faster than large-v3.

*ThinkCentre mini PC, 4-core i5-6500T, 1 transcription thread. †Ryzen 9 3900X limited to 4 and 20 transcription threads; a slower CPU will be slower. ‡RTX 2070 SUPER. Without language detection (adds ~80 %). Short recordings are the slower case per second of audio: nine longer ones (13 s to 7 min) ran up to 2× faster (the mini PC: small 0.44×, medium 1.0×), but needed more memory (medium peaked at 3.2 GB on the 7-minute one). Rough figures, one run each.

Installation

One-line installer (Linux)

Detects your distribution and installs the matching package:

curl -fsSL https://raw.githubusercontent.com/rafafunt/FalconSDR/master/packaging/install.sh | bash

Debian/Ubuntu get the .deb, Fedora/openSUSE the .rpm, Arch the AUR package, anything else the AppImage. Add --from-source to install from a git checkout instead: the way to get SoapySDR support (LimeSDR, SDRplay, USRP, …, and HackRF/Airspy through SoapySDR instead of the built-in access) and the fallback for other CPU architectures or musl distros such as Alpine. The built-in HackRF/Airspy access needs their system libraries, which a source install does not bring along: add --with-sdr-libs (installs libhackrf0 and libairspy0 with apt, or hackrf and airspyone_host with dnf, using sudo): curl -fsSL … | bash -s -- --from-source --with-sdr-libs.

Ubuntu / Debian - .deb package

Download the .deb from Releases and install it:

sudo apt install ./falconsdr_<version>_amd64.deb

Start FalconSDR from the application menu. Unplug and replug the dongle once after installing - the package adds a udev rule for USB access and stops the DVB-T TV driver from taking the dongle.

This package is self-contained (RTL-SDR, plus HackRF and Airspy natively - experimental; no Python needed). For SoapySDR devices, run from source instead.

Fedora / openSUSE - .rpm package

Download the .rpm from Releases and install it:

sudo dnf install ./falconsdr-<version>-1.<arch>.rpm

Same self-contained bundle and the same udev/DVB-T handling as the .deb; replug the dongle once after installing. Needs glibc 2.38+ (current Fedora / openSUSE Tumbleweed); RHEL/Rocky 9 are too old for this build - run from source there instead.

Arch Linux - AUR

yay -S falconsdr

(repackages the official release build; maintained by @9M2PJU).

Any distribution - AppImage

Download FalconSDR-<version>-<arch>.AppImage from Releases, make it executable and run it - no install, no root:

chmod +x FalconSDR-*.AppImage
./FalconSDR-*.AppImage

One manual step on a machine with no RTL-SDR software installed yet: copy packaging/linux-deb/60-rtl-sdr.rules to /etc/udev/rules.d/, reload udev and replug the dongle (the .deb/.rpm do this for you).

Windows - installer or portable ZIP

Download from Releases either FalconSDR-<version>-windows-x64-setup.exe (installs for your user, no administrator rights needed) or FalconSDR-<version>-windows-x64-portable.zip (unzip anywhere and run falconsdr.exe). Both are self-contained - no Python needed; MP3 recording works out of the box and transcription installs from Settings like in any other install. The program is not code-signed yet: if SmartScreen says "Windows protected your PC", choose More info → Run anyway.

RTL-SDR driver: Windows needs the WinUSB driver on the dongle, installed once with Zadig. If SDR++, SDR# or another SDR program already works with your dongle, it's already done. Otherwise: plug in the dongle → run Zadig → Options → List All Devices → pick "Bulk-In, Interface (Interface 0)" (or "RTL2838UHIDIR"; not Interface 1) → target driver WinUSB → Replace Driver → replug the dongle. The same steps are in the app (Listen → SDR device). A HackRF or Airspy may need the same driver step; the built-in access to them is untested on Windows with real hardware.

FalconSDR runs in the background with its interface in the browser; quit it in Settings → Advanced. "Start with the system" is there too (off by default).

macOS - experimental

Download FalconSDR-<version>-macos-arm64-experimental.zip (Apple Silicon: M1 and newer) or …-macos-x86_64-… (Intel Macs) from Releases, unzip it and move FalconSDR to Applications. These builds are made automatically and pass the same self-test as the other packages, but haven't been tried on a real Mac with an SDR yet - feedback in Discussions is very welcome.

The app isn't signed by Apple. The first time, right-click it → Open (or System Settings → Privacy & Security → Open Anyway). If macOS says the app "is damaged", that's the same missing signature; clear the download flag in Terminal:

xattr -dr com.apple.quarantine /Applications/FalconSDR.app

An RTL-SDR needs no extra driver on macOS. Builds since 1.4.1 also carry the HackRF and Airspy libraries (built-in access, experimental); they only check that the libraries load - no real Mac with a HackRF or Airspy has tried them.

From source (Linux, Windows, macOS)

Requires Python 3.10+.

git clone https://github.com/rafafunt/FalconSDR.git
cd FalconSDR
python3 -m venv --system-site-packages .venv       # lets the venv see python3-soapysdr (Linux)
.venv/bin/pip install -r requirements.txt        # Windows: .venv\Scripts\pip ...
.venv/bin/python main.py                          # Windows: .venv\Scripts\python main.py

The interface opens at http://127.0.0.1:8765. To update: git pull, run the pip install line again and restart (or run the one-liner with --from-source again from the folder above FalconSDR/).

• Linux: for RTL-SDR access without root, install your distribution's rtl-sdr package (udev rules) or copy packaging/linux-deb/60-rtl-sdr.rules to /etc/udev/rules.d/. For SoapySDR devices install python3-soapysdr and the device's soapysdr-module-* (e.g. sudo apt install python3-soapysdr soapysdr-module-hackrf); the venv must be created with --system-site-packages, as above, or it cannot see them. HackRF and Airspy can also be opened without SoapySDR (experimental, SDR device → Access method; the Windows, .deb, .rpm and AppImage builds already carry the libraries): sudo apt install libhackrf0 libairspy0, or run the one-line installer with --with-sdr-libs.
• Windows: the dongle needs the WinUSB driver - see RTL-SDR driver above.
• MP3 recordings use the ffmpeg that comes with the imageio-ffmpeg package (or a system ffmpeg on PATH); on platforms without a wheel (32-bit ARM) install ffmpeg from the system, otherwise recordings stay WAV.

Updating and removing

Quit FalconSDR, then install the newer .deb, .rpm or Windows installer over the old one - settings, channels, favorites and recordings are kept. Start it again afterwards (a copy left running through the update shows errors). An AppImage is replaced by downloading the new one. A source install is updated by running the --from-source line again from the folder that holds FalconSDR/ (it pulls the checkout and refreshes the venv), or by git pull in the checkout and the pip install line; the headless receiver, by its own line again (see below). Remove with sudo apt remove falconsdr, sudo dnf remove falconsdr or Windows Settings → Apps; your recordings and settings stay on disk (see Where things are stored).

Headless receiver

On a Linux machine with the SDR attached (Debian, Ubuntu, Raspberry Pi OS):

curl -fsSL https://raw.githubusercontent.com/rafafunt/FalconSDR/master/packaging/install.sh | bash -s -- --server

It installs the latest release as a service that starts on boot and ends with the address to open from any device on your network (http://<receiver-ip>:8765). Run the same line again to update - settings and recordings are kept. Details, other versions and helper computers for transcription: packaging/remote-server/README.md. It also installs libhackrf0 and libairspy0 (small), so a HackRF or Airspy works through the built-in access without extra steps. The service sees the system's SoapySDR: sudo apt install python3-soapysdr soapysdr-module-<your receiver> and restart it (sudo systemctl restart falconsdr-receiver), then pick the access method under SDR device.

First steps

1. Listen page: pick your SDR, then add a band (or a preset) and press Start.
2. Raise or lower the trigger level until real transmissions open and noise doesn't.
3. Switch on recording when you're happy with what it stops on.
4. Name the frequencies you hear in the Channels page (or straight from a recording) - names show up everywhere, including search.

Tips for good recordings

• Trigger level decides what counts as a transmission. Lower = weaker stations get recorded, but more noise too.
• Some frequencies are just a constant carrier or interference (birdies from your own computer or USB hub, a nearby transmitter's harmonics, a dead carrier). They trip the trigger all the time and keep you from lowering it. Add them to the blacklist (Listen → Skip + blacklist, or Channels → Blacklist). After that you can usually lower the trigger level by several dB and pick up the weak stations you were missing.
• Gain defaults to Auto (software). The tuner's own hardware AGC tends to overload on a strong nearby station, which shows up as false detections on neighbouring channels.
• Hold after a transmission (5 s by default) keeps a recording open through the pause between a call and its answer.
• A narrower step in a band means more channels to check per sweep; use the channel spacing the band really uses.
• If the Activity feed says this computer can't keep up with the SDR, data is being lost (short gaps in audio and recordings): lower the capture bandwidth (a wide HackRF capture costs much more than an RTL-SDR's 2.4 MHz), or move off a virtual machine - USB passed through to a VM is a common cause.

Command line

falconsdr for the installed packages (Windows: falconsdr.exe in its install folder), python main.py from source:

falconsdr               local mode: scan on this computer, UI on 127.0.0.1:8765
falconsdr --server      reachable from the whole network
falconsdr --start       start scanning right away
falconsdr --local       scan locally even if this computer is set up as a client of a server
falconsdr --port 9000   another port
falconsdr --no-browser  don't open the browser
falconsdr --version     show the version
falconsdr --self-test   check this install has everything it needs
falconsdr --probe-sdr   measure the receiver with the scanner stopped (see Hardware)

Where things are stored

What	Where
Settings, channels, blacklist	~/.config/FalconSDR/
Recordings (default, changeable)	~/FalconSDR/recordings/, one folder per day
Quality scores, transcripts, favorites	~/.local/share/FalconSDR/

(~ is your home folder, also on Windows.)

Troubleshooting

• "Failed to connect… retrying" - the SDR is missing, in use by another program, lacks USB permissions (Linux: udev rule, replug) or has the wrong driver (Windows: Zadig).
• A HackRF or Airspy is plugged in but not listed - on Linux the Listen page names what is missing: SoapySDR (source install; see above) or the libhackrf / libairspy library (sudo apt install libhackrf0 libairspy0, Fedora: sudo dnf install hackrf airspyone_host, or the installer's --with-sdr-libs). If one way to open a receiver fails, try the other under SDR device → Access method.
• Nothing gets recorded - recording is off, the trigger level is too high, or the band's modulation is wrong (aviation is AM, most other VHF/UHF voice is NFM).
• Constant recordings of noise - blacklist the offending frequencies (see above) or raise the trigger level.
• Distorted audio on strong stations - switch gain to Auto (software) or lower the manual gain.

Feedback and contributing

• Questions, ideas, your setup: Discussions
• Bugs: Issues - please include the version (Settings → Advanced, at the bottom), your OS, how you installed it, your SDR and the band.
• Translations: the interface is English with a Polish translation in falconsdr/locale/pl.po; a new language is a copy of that file (see falconsdr/i18n.py).
• Code: pull requests are welcome; AGENTS.md describes the architecture and the pitfalls already hit once. Tests: python -m unittest tests/test_core.py.

Legal

The laws on radio monitoring differ a lot between countries. In some places you may listen to most bands but not record or share what you hear; in others even listening to non-public services (aviation, police, emergency services, military) is forbidden. Check what applies where you live before you scan, and especially before you record or publish anything - you alone are responsible for how you use this program.

License

GPL-3.0-or-later (see LICENSE) - required by pyrtlsdr and the bundled librtlsdr. The Windows, Linux and (since 1.4.1) macOS packages list every bundled component and its license in THIRD-PARTY-LICENSES.txt (ffmpeg in them is a GPLv3 build; libhackrf and libairspy are BSD-3-Clause, libusb LGPL-2.1). Transcription software and Whisper models are not bundled: they are downloaded on the user's own request from their publishers (faster-whisper, CTranslate2 and the default model Systran/faster-whisper-large-v3 are MIT-licensed, as is the optional English aviation model jacktol/whisper-medium.en-fine-tuned-for-ATC-faster-whisper). Neither is the digital voice decoder: on request, the app downloads dsd-fme and mbelib from their authors (the authors' Windows build; source compiled on the user's machine elsewhere) and runs dsd-fme as a separate program.