AUDIO STEGANOGRAPHY BASICS

Category: OSINT, Forensics, Audio Steganography

INTRODUCTION

Hello everyone.

This document is an introduction to audio spectral analysis.

The objective of this course is to get you comfortable with the basic concepts of spectrograms in the context of OSINT.

The course will be divided into three main parts:

• Introduction to waveforms and their limits
• Introduction to spectrograms and their advantages
• Introduction to audio steganography using spectrograms

The overall structure of the course assumes that you are already familiar with general steganography concepts and their OSINT applications. If this is not the case, I highly recommend looking into them before proceeding.

Let’s dive in.

WAVEFORMS

You’ve probably already opened an audio file (.mp3, .wav, etc.) in a sound analyzer, such as Audacity.

In addition to reading audio files, these programs offer the ability to visualize the structure of the sound.

The first and most known visual representation of a sound you may encounter is called a waveform.

A waveform is a 2D representation of the sound’s amplitude over time, where X = Time and Y = Amplitude, as shown here:

Although waveforms provide some good insight into the sound’s composition, they leave some important questions unanswered. Most importantly, in OSINT contexts, sound frequencies are often worth analysing. This is when spectrograms step in.

SPECTROGRAMS

There are many open-source sound analyzers available. I personally use Sonic Visualiser available on Windows, Linux and MacOS. Examples in this course will be made with it.

Let’s take a look at how spectrograms look compared to waveforms. For this example, we will use the .mp3 of the song “Billie Jean” by Michael Jackson.

The upper graph, in blue, shows the waveform as we’ve seen before. Just as we’ve discussed, its focus is on Amplitude (y) over time (x).

While the spectrogram’s x axis remains the same (Time), the first difference dwells in the y axis. In the lower (green) graph, let’s take a closer look at the scale on the left side:

The scale is divided in two parts. The colored scale on the left represents the relative strength of each frequency, here expressed in decibels (dB), which we’ll discuss in a minute. The right part represents the frequencies measured in hertz (Hz).

These frequencies, on the graph, belong to the y axis. Hence, while on the waveform, y/x measured the evolution of Amplitude over Time, on the spectrogram y/x = Frequency/Time.

But the real power of spectrograms is revealed when we add their third dimension, which we will call the z axis to ease the explanations.

Please note that, although it is referred to as an ‘axis’ or ‘dimension’, its representation in the spectrogram is neither physical nor spatial, but rather uses colors.

This dimension represents the strength (z) of each frequency (y), at a given moment (x).

As indicated on the above scale, each color represents a given decibel value, or strength.

To better understand the difference between the strength value of the spectrogram and the amplitude value of the waveform, let’s have a look at Billie Jean’s intro:

Here, you may notice that as we move along the x axis, the amplitude peaks in the waveform correspond to distinguishable yellow bands in the spectrogram. Although strength and amplitude are different measures, they are closely related: a strong increase in the waveform’s amplitude will generally correspond to stronger frequency components in the spectrogram.

The difference between the two is explained by the fact that the waveform represents the total amplitude of a signal (you can think of it as all frequencies of the spectrogram merged as one).

In simple words, the difference between the two can be expressed as such:

“While the waveform tells us when the whole signal gets stronger, the spectrogram tells us where that strength is distributed among the frequencies.”

That said, you already know enough to dive into the basics of audio steganography.

SPECTRAL STEGANOGRAPHY

Since spectrograms are visual representations of sound in three dimensions, and since they use colors to differentiate levels of strength, it’s time to introduce the possibility of “drawing” specific shapes inside a spectrogram by adding specific strength variations at specific times and within specific frequency ranges.

Let’s use a “semi-real-life” example.

“Elite Dangerous” is a video game that uses spectral steganography to unveil some of its mysteries and entertain players. Within its universe, different places and objects sometimes emit “strange” signals that barely mean anything if you only listen to them. However, through spectral analysis, some of these may reveal hidden messages and unlock new hints to important questions around the game.

When I first heard about this, I was barely starting my OSINT journey and didn’t expect to write about it someday. Time went by, and as I began writing this course, I searched the whole internet for the audio recordings that I’d once heard. Unfortunately, the exact recording I was looking for dates back to 2017 and can no longer be found online. However I found a more recent audio recording using the same technique, and we’re going to use it now.

As I mentioned, the audio signal barely means anything to the ear, but its spectral analysis reveals some interesting shapes:

At first glance, it doesn’t mean anything but after adjusting the frequency range and some display settings, we get a much clearer picture:

Inside a simple .mp3 file, somebody “drew” what seems to be human shapes.

Because this course is just an introduction, we will not discuss the setting adjustments here. Neither will we mention the technical means to insert such shapes in an audio file. However, you may begin to understand that any type of message could be hidden in a spectrogram using this method.

Whether it’s a drawing, a clear text or a code, anything that can be written down physically can be encoded into an audio signal.

All one needs to do is insert the right frequencies, at the right strengths, at the right times, and let the spectrogram speak.

CONCLUSION

In OSINT, the applications of steganography are legion. You’re probably already familiar with steghide, which can embed hidden data inside ordinary files.

Consider the method introduced here as a similar way to hide messages inside a sound in any audio file.

Thanks for reading.