Signal Subspace Speech Enhancement with Oblique Projection and Normalization

Abstract

In this paper, a subspace speech enhancement method handling colored noise using oblique projection is proposed. Perceptual features and variance normalization are used to reduce residual noise and improve speech intelligibility of the output. Initially, additive noise is removed from the noisy speech by removing the orthogonal noise subspace from the noisy speech subspace to obtain the speech subspace. Then, the oblique projection of the noise subspace on the speech subspace along the additive noise subspace is used to determine the colored noise that remains. The enhanced clean speech signal is estimated using Spectral Domain Constrained Estimator, incorporating the masking property of the auditory system and the variance of the colored noise. To avoid the occurrence of any abrupt spikes in the output, variance normalization is performed by adaptively changing the control parameter of the gain matrix. The spectrogram, objective measures and subjective intelligibility test show superior performance of the proposed method over other existing speech enhancement methods.

Keywords

• Speech enhancement
• signal subspace approach
• masking property
• oblique projection
• variance normalization