Bark filter approach[10]

Vetter replaces the eigenfilters in the subspace method by Bark filtering to make use of the prior knowledge held about the human auditory system. For maximum energy compaction, the filtering is processed in the square DCT domain.

$$X(k)_{Bark}=\sum_{j=-b/2}^{b/2} G(j,k){X(k)}^2 \quad k=0,\ldots,N-1$$

where $b+1$ is the processing width of the filter, $G(j,k)$ is the Barkfilter whose bandwidth depends on $k$ and $X(k)$ are the DCT coefficients.

Noise masking is achieved in the human auditory system because it cannot distinguish two signals close in time or frequency domains. The critical filterbank known as the Bark filterbank achieves simultaneous masking. Perception of a signal at a particular frequency is influenced by the energy of a perturbing signal in a critical band around this frequency. The bandwidth of the critical band varies with frequency, beginning at around 100 Hz below 1 KHz and then increasing up to 1 KHz above 4 KHz (decreased sensitivity for higher frequencies). The Bark filterbank gives equal weight to portions of speech with the same perceptual information.