An audio apparatus is suitable for generating crowd sounds from an audio signal is disclosed in which the apparatus comprises modulation means operable to modulate a noise signal in response to the audio signal to generate a modulated noise signal, and diffusion delay means. The diffusion delay means is operable to apply a series of two or more delay operations, the input signal to a first such delay operation in the series being the modulated noise signal, and input to each subsequent delay operation in the series being the output signal generated by a preceding delay operation. Each delay operation comprises modifying that operation's input signal by the addition of a delayed version of that operation's input signal.