Reconstructing printed text solely based on the sound emitted during
printing was the major goal of my master’s thesis. In this talk, I
will present a fully automated approach for reconstructing printouts
from dot-matrix printers. The approach makes use of many ideas from
speech recognition and other audio signal processing areas.
Furthermore, I will show that the attack can be conducted with cheap,
off-the-shelf equipment. The quality of the results is influenced by
several parameters, e.g., by the precise printer settings and by
environmental settings such as the noise shielding incorporated in
modern printers.