Centre for

Speech Enhancement Tutorial - Noise Estimation

4.2 Stationary Noise Model

The most common model of the noise is that it is a Gaussian process with a slowly changing power spectrum. The spectrum is normally represented by individual spectral, mel-spectral or cepstral coefficients. In Ephraim et al. [1989a] an all-pole spectrum model is used for the noise process and an approximation to the maximum likelihood estimate of the AR parameters is provided by conventional LPC analysis.

Previous | Next