Image to Sound — Sonification, Spectrogram Art and vOICe

Most popular "image to music" tools online today (InsMelo, Mubert, Suno, Artlist) generate a full song from your photo using AI — they read the mood and palette and compose a track with melody and drums. This tool does sonification instead: it maps pixel data directly to sound. Musical mode adds a scale so the result feels like music, but it is still a 1:1 transformation of your image, not a composed song. For AI-composed music, look for tools labeled "AI music from image".

Switch the source to Text, type up to three lines, and download the WAV. The word is drawn into the frequencies, so it is invisible in a waveform and plain to see in a spectrogram view — Audacity (Spectrogram view), Spek, Sonic Visualiser or any phone spectrum-analyzer app. One choice decides whether it survives: under Frequency range, "Survives sharing" puts the text in a narrow band below the cutoff MP3 and AAC apply, so a messenger or a video site re-encoding the file does not erase it. Outline style reads more sharply than filled at low detail.

Musical mode reads each time-slice as a brightness profile, folds it onto the degrees of your scale and sounds every degree that stands out — up to eight at once. Melodic, and it changes when the picture changes. Raw mode is additive synthesis: every pixel becomes a sine partial whose frequency comes from its row and whose volume comes from its brightness, so the audio is literally the spectrogram of your image — harsh and drone-like, and the mode for spectrogram art. vOICe mode is the same idea held to one exact specification, matching the sensory-substitution system of that name.

vOICe mode is the one to use, and the live camera scan is what makes it a device rather than a converter: the camera is read continuously, left to right, in the mapping blind users actually train on — 500–5000 Hz on an exponential axis, sixteen brightness steps two decibels apart with black as true silence, a stereo fade so the ear can tell left from right, and the click that marks where each scan begins. Headphones matter: without them the stereo fade, which is half of what makes a soundscape readable, is lost. None of this replaces a screen reader or alt text; it adds a way to perceive shape and layout through sound.

Ones with strong contrast and few large shapes. In musical mode a face, a simple object or a line of text produces a melody you can recognize again; in raw mode geometric patterns, diagonals and gradients give the most distinctive spectrogram textures, since the sound is quite literally the shape of the image. What produces almost nothing is the opposite: a very dark photograph has no bright regions to sound, and a uniform one — a wall, a sky, a blurred background — has nothing that changes, so the output drones. The Contrast slider under "Picture and color" is there for exactly this, and the frame on screen shows what it did before you listen.

Detail sets the number of rows, which is the frequency resolution: 32, 64, 128 or 192. The number of columns follows the shape of your picture rather than squashing it into a fixed grid, and the frame shown on screen is the one the engine reads, dimensions and all. More rows mean a sharper picture in the spectrogram and a longer render; 64 is enough for text and shapes, 128 and above for photographs.