Nonlinear diffusion has long proven its capability for discontinuity-preserving removal of noise in image processing. We investigate the possibility to employ diffusion ideas for the denoising of audio signals. An important difference between image and audio signals is which parts of the signal are considered as useful information and noise. While small-scale oscillations in visual images are noise, they encode essential information in audio data. To adapt diffusion to this setting, we apply it to the coefficients of a wavelet decomposition instead of the audio samples themselves. Experiments demonstrate that the denoising results are surprisingly good in view of the simplicity of our approach. Nonlinear diffusion promises therefore to become a powerful tool also in audio signal processing.