Sonic

Send text and files through sound. Play it on one device, listen on another, and the data arrives byte-for-byte or not at all.

• Send: type a message or drop a file, press Play.
• Receive: press Listen. It picks up the transmission and decodes it on its own.
• Explore: see the sound itself: waveform, spectrum, spectrogram, the tones and how they separate.
• Every transfer is checked against a SHA-256 fingerprint.
• Works across a desk (tested laptop → phone at half a metre). Not yet across a room.

Run it

Requires Python 3.12+ and Node 20+.

cd backend && pip install -e ".[all]" && sonic serve       # backend on :8000
cd frontend && npm install && npm run dev                  # app on http://localhost:5173

To receive on a phone on the same Wi-Fi, start the app with SONIC_LAN=1 npm run dev and open https://<your-computer's-ip>:5173/#receive on the phone. Phones only allow the microphone over HTTPS, so accept the self-signed certificate warning.

There is also a command line:

sonic encode notes.txt -o notes.wav     # file → sound
sonic decode notes.wav -o notes.txt     # sound → file
sonic play notes.wav                    # through the speakers
sonic record --seconds 10 -o rec.wav    # from the microphone

---

How it works

data → compress → packets + checksums → error correction → tones → speaker
mic → find the start → separate the tones → correct errors → verify SHA-256 → data

1. Pack. The data is compressed, split into packets with a CRC-32 each, and protected with a convolutional error-correcting code.
2. Play. The bits are spread over hundreds of tones sounding at once between 2 and 10 kHz (OFDM, the technique Wi-Fi uses on radio). Each tone carries a few bits per symbol (BPSK to 256-QAM). A short chirp marks the start, and a fixed, extra-robust header describes the rest.
3. Rebuild. The receiver finds the chirp, measures how the room and devices changed each tone, undoes it, corrects errors, and checks every packet and the final SHA-256. Anything that doesn't match is rejected.

Speeds

Speed	Encoding	Use it for
Fast	16-QAM	Devices a hand's width apart
Balanced	QPSK	About half a metre (tested laptop → phone)
Reliable	BPSK, long symbols	Echoey rooms; slowest

What we measured

One MacBook Pro and one phone in an ordinary room:

Setup	Result	Signal quality
Laptop speaker → its own mic	Received, even at 64-QAM	36 dB
Laptop → phone, 0.5 m	Received	21.5 dB
Laptop → phone, 2 m	Not received	3 dB after room echoes

At a distance, reflections off the walls arrive later than each symbol can absorb and blur into the next one. Louder doesn't help. That's the main thing to solve next.

Limits

• Audio carries only a few kilobits per second, so this suits messages, links and small files, not photos.
• Decoding runs in the Python backend, so the receiving page needs a connection to it.
• Low-bitrate voice codecs (e.g. a 32 kbit/s voice note) destroy the signal; MP3 and AAC survive.
• No encryption: anyone in earshot can receive.

More detail

• docs/protocol.md: frame format, header, packets
• docs/architecture.md: code layout
• docs/research_log.md: every experiment and measurement, including the failures
• docs/limitations.md: known limits

backend/    Python engine, CLI, API and tests (cd backend && pytest)
frontend/   web app
docs/       protocol, architecture, research log