Abstract:A Bark-band residual noise model integrated with the human hearing mechanism is proposed to efficiently complement sinusoidal model in parametric audio coding. The time-varying spectrum of the residual noise is retrieved by Bark-scale piecewise constant magnitude estimates along with random phases. In the proposed noise model, Bark bands information is obtained by short-time FFT method and window overlap-add technique is exploited to remove boundary discontinuities. SVQ is also incorporated into parameter quantization process for the low bit-rate coding demand. Simulation results and informal listening tests show that when the sinusoidal model is combined with the Bark-band noise model, better synthesis audio quality can be achieved compared with the original sinusoidal modeling audio codec.

Bark-Band Residual Noise Model for Parametric Audio Coding