We present a statistical method for restoring audio signals which have been coarsely quantized without dither. The signal is modeled as a piecewise-stationary linear autoregressive process. Parameter and signal estimation are performed using sampling methods. To improve performance, sinusoidal components are incorporated into the model. Model order estimation and sinusoid selection are found to be crucial to the quality of the restoration.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.