SPECTRAL

Dennis Jegel

HD Video + Audio 2:05 min

SPECTRAL is an audiovisual analysis of the measurable differences between human and artificial voice. Through spectrogram representation, the frequency patterns of both voice types are contrasted as visual data. What appears acoustically similar reveals its different structure in spectral analysis: The AI voice as technical reconstruction, the human voice as carrier of emotion and experience. The work visualizes what lies between original and copy – those qualities that elude digital capture.

The making of SPECTRAL

The project originated from an investigation into audio security and hidden data injection – the possibility of hiding information within sound files that cannot be heard. This technical starting point led me to explore spectrogram analysis as a method of making the invisible visible, revealing what lies beneath the surface of audio.

Initially, I experimented with creating animated spectral representations – essentially attempting to build a flipbook-style animation by rapidly sequencing spectrogram frames within audio software. However, this approach felt too broad and technically complex without a clear conceptual anchor. The breakthrough came through collaboration with Lukas, whose focus on synthetic voices opened a new direction for my spectral analysis work.

While Lukas explored AI-generated voices from a conceptual perspective, I became interested in how my technical approach could contribute to this field. This led me to investigate a compelling research question: if AI can replicate human speech convincingly to the ear, what does spectral analysis reveal about the fundamental differences? This became the core of SPECTRAL – using technical visualization to expose what acoustic similarity conceals.

The visual approach centers on direct data representation. Rather than metaphorical imagery, the spectrogram itself becomes the primary visual element – raw frequency data translated into visual patterns. The work presents both human and AI voices as spectral data, allowing the technology to reveal subtle structural differences that may not be immediately apparent to the eye.

The audio layer features the recorded dialogue between human and AI voices, creating a direct comparison that visitors can both hear and see simultaneously. The spectral visualization allows viewers to observe the frequency signatures of identical-sounding words.

SPECTRAL ultimately demonstrates that while AI can achieve acoustic similarity, the spectral domain offers a technical method for examining the differences between authentic human expression and its digital reconstruction.