Speckle-interference-based photon-level audio reconstruction

Remote acoustic sensing under ultra-low photon flux is challenging because conventional methods rely on either stable reference arms or high illumination levels. Here, we demonstrate a reference-free scheme for audio reconstruction from speckle-field photon statistics. Acoustic microvibrations are encoded into temporal intensity modulations of scattered light, recorded as sparse photon timestamps, and reconstructed through Gaussian temporal kernelization, windowed integration, and progressive denoising. In a free-space experiment, intelligible speech is recovered over 11–522 kcps. Rhythm and pauses remain identifiable at 11 kcps, while automatic speech recognition accuracy reaches 84% at 522 kcps. This approach reduces the required optical power to the femtowatt level, enabling robust acoustic sensing in photon-starved environments.