Image2Spectrogram 🎵🖼️

Hide an image inside an audio file. Open it in a spectrogram viewer and the picture appears.

---

🎉 After one years in beta, Image2Spectrogram has finally graduated and shipped its first stable release.

Overview

Image2Spectrogram is a small command-line tool that converts an image into a .wav file. The audio is synthesized so that its spectrogram (frequency over time) draws the original image.

Each image column becomes a short slice of audio, and each row becomes a sine wave at a specific frequency. Dark pixels switch their frequency on, light pixels leave it silent. Played back, it sounds like a burst of tones; viewed in a spectrogram, it reads as a picture.

image (pixels)  ──►  columns = time  ──►  rows = frequency  ──►  output.wav

Features

• Simple CLI: pass an image, get a WAV. Every parameter is configurable.
• Clean audio: a Hann window is applied to every column to prevent clicks and pops at column boundaries.
• High-quality resizing: images are downscaled with Lanczos resampling to preserve detail.
• Resolution guard: warns you when the chosen height is finer than a spectrogram viewer can resolve.
• Standard output: 16-bit PCM WAV, 44.1 kHz by default.

Requirements

• Python 3.9 or newer
• NumPy, SciPy and Pillow
• A spectrogram viewer, such as Audacity

Installation

git clone https://github.com/SirAtilotty/Image2Spectogram.git
cd Image2Spectogram/Image2Spectrogram

Create a virtual environment and install the dependencies:

python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt

Arch Linux and other distros with a managed Python: pip install outside a virtual environment fails with externally-managed-environment (PEP 668). Use the venv above, or install the system packages with sudo pacman -S python-numpy python-scipy python-pillow.

Usage

python spectrogram_text.py message.png

This writes output.wav to the current folder. Open it in Audacity, click the track name, choose Spectrogram, and your image shows up.

Examples

# Custom output file
python spectrogram_text.py logo.png -o logo.wav

# Higher resolution (longer audio, finer detail)
python spectrogram_text.py logo.png --width 256 --height 96

# Wider frequency range and a lower tone threshold
python spectrogram_text.py logo.png --freq-start 200 --freq-end 10000 --threshold 0.4

Options

Option	Default	Description
image	required	Path to the source image.
-o, --output	output.wav	Output WAV file path.
--width	128	Image width in pixels after resizing. More width means longer audio.
--height	64	Image height in pixels after resizing. More height means finer detail and does not change the duration.
--sample-rate	44100	Audio sample rate in Hz.
--duration-per-column	0.05	Seconds of audio per image column.
--freq-start	300	Lowest frequency in Hz (bottom row of the image).
--freq-end	8000	Highest frequency in Hz (top row of the image).
--threshold	0.5	Minimum pixel intensity (0 to 1) that triggers a tone.
--fft-size	2048	FFT window of your spectrogram viewer. Only used for the resolution warning.

Run python spectrogram_text.py -h for the full help text.

Preparing your image

• Dark on light. Dark pixels produce tones and appear bright in the spectrogram, so use a dark subject on a white background.
• High contrast. Thick shapes, silhouettes and bold text survive downscaling. Thin lines and fine gradients get lost.
• Match the aspect ratio. The image is resized to --width x --height, so use the same ratio (the default 128x64 is 2:1) to avoid stretching. The source file can be any size at least as large as the target.
• Keep the height reasonable. Rows closer together than the viewer's frequency resolution blur into each other. With the defaults, about 358 rows is the practical ceiling, and the tool warns you if you go past it.

How it works

1. The image is converted to grayscale, resized with Lanczos resampling, and inverted so dark pixels carry high intensity.
2. Columns map to time: each one lasts --duration-per-column seconds.
3. Rows map linearly to frequency between --freq-start and --freq-end.
4. For every column, a sine wave is added for each row whose intensity exceeds --threshold.
5. Each column is multiplied by a Hann window, concatenated, normalized and written as 16-bit PCM.

Total duration is width x duration-per-column, so the default 128 columns produce about 6.4 seconds of audio.

Troubleshooting

Problem	Fix
error: the following arguments are required: image	Pass the image path: python spectrogram_text.py message.png.
externally-managed-environment	Use a virtual environment (see Installation).
Signal is silent	The image produced no tones. Check that the subject is dark on a light background, or lower --threshold.
Image looks blurry or smeared in Audacity	Lower --height, or raise the window size in Audacity under Spectrogram Settings.
Image looks squashed	Make the source image match the aspect ratio of --width x --height.

Roadmap

• Optional logarithmic frequency scaling
• Preview of the resized image before synthesis
• Unit tests and CI

License

Released under the MIT License. See LICENSE for details.

Copyright (c) 2026 Atilla İlhan.