AudioImage

An acoustic modem that transmits images over sound. An image is encoded into an audio signal, played through a speaker, recorded by a microphone on another device, and reconstructed on the receiving end.

Transmission is based on frequency-shift keying (FSK), with the receiver reading data via the FFT. The pipeline includes chirp-based synchronisation, a self-describing packet header, optional DCT compression (grayscale and color), and Reed-Solomon forward error correction.

Features

• FSK modulation with FFT demodulation
• Chirp preamble synchronisation via cross-correlation
• Six raw image formats (RGB888, RGB565, RGB332, GRAY8, GRAY4, BW1)
• DCT compression (grayscale and YCbCr 4:2:0 color)
• Reed-Solomon error correction
• Parallel transmission (multiple bytes per window)
• Transmit to a WAV file or play directly through a speaker
• Receive from a WAV file or record live from a microphone

Requirements

• Python 3.10+
• numpy
• scipy
• pillow
• sounddevice
• reedsolo

Install with:

pip install numpy scipy pillow sounddevice reedsolo

Usage

Sending

python send.py <image> <output.wav | --play> [options]

Encode an image to a WAV file:

python send.py images/test.png out.wav

Play directly through the speaker instead of writing a file:

python send.py images/test.png --play

Receiving

python receive.py <input.wav | --record <seconds>> <output_image> [options]

Decode from a WAV file:

python receive.py out.wav recovered.png

Record live from the microphone for a given number of seconds:

python receive.py --record 10 recovered.png

Options (send.py)

Option	Description	Default
--format F	Raw image format, F from 0 to 5 (see below)	0 (RGB888)
--dct Q	Grayscale DCT compression at quality Q (1–100)	off
--dct-color Q	Color DCT compression (YCbCr 4:2:0) at quality Q	off
--compress	Run-length encoding (RLE)	off
--fec N	Reed-Solomon with N ECC bytes, corrects up to N/2 errors per block	0 (off)
--parallel N	Send N bytes per window in parallel (faster, wider band)	1
--spec PATH	Save a spectrogram of the signal to PATH	off
--show-spec	Display the spectrogram interactively	off

Image formats

Value	Format	Bits/pixel
0	RGB888	24
1	GRAY8	8
2	RGB332	8
3	RGB565	16
4	GRAY4	4
5	BW1	1

The header records the format and all other parameters, so the receiver needs no options to match; it reads everything from the transmitted header.

Options (receive.py)

Option	Description
--spec PATH	Save a spectrogram of the recorded signal to PATH
--show-spec	Display the spectrogram interactively

Examples

Color DCT at quality 50 with error correction, written to a file:

python send.py images/test.png out.wav --dct-color 50 --fec 32
python receive.py out.wav recovered.png

Live transmission over the air (run on two machines):

# Receiver (start first)
python receive.py --record 15 recovered.png

# Sender
python send.py images/test.png --play --format 0 --fec 32

Project structure

send.py            Encoder entry point
receive.py         Decoder entry point
modem/             Core library
  protocol.py        Constants (sample rate, window size, frequencies)
  encoder.py         FSK modulation
  decoder.py         FFT demodulation
  sync.py            Chirp generation and cross-correlation
  header.py          Self-describing packet header
  image_codec.py     Image loading and raw format packing
  dct_codec.py       DCT compression (grayscale and color)
  compress.py        Run-length encoding
  fec.py             Reed-Solomon error correction
  audio_io.py        WAV I/O, playback, recording
  specplot.py        Spectrogram plotting
sweep/             Frequency-response measurement tools
  sweep_send.py      Generate and play a frequency sweep
  sweep_record.py    Record the sweep
  sweep_analyze.py   Analyse the recorded sweep (frequency response)

Configuration

Core parameters live in modem/protocol.py:

• FS — sample rate (44100 Hz)
• N — samples per window (294; larger means slower but finer frequency resolution)
• BASE_FREQ, SPACING_HZ — carrier band layout
• IMG_SIZE — default image size images are scaled to (32)