| Summary: | Speech enhancement aims to improve the performance of speech processing systems
operating in various noisy environments. The performance of speech enhancement
algorithm can be evaluated by two uncorrelated criteria: clarity and intelligibility.
Here, we present a speech enhancement algorithm based on the signal subspace
method, which can be adopted for arbitrary noise types. Firstly, an equalizer is
introduced to whiten the noise embedded in the noisy speech signal. Then, by
applying the Karhunen-Loeve transform (KLT), the noisy speech signal is
decomposed into two subspaces: noise subspace and signal-plus-noise subspace. The
clean signal can be estimated from the signal-plus-noise subspace after eliminating
the noise subspace. By assuming the noise is additive and uncorrelated with clean
signal, the recovery of the original signal is conducted frame-by-frame by
introducing two criteria: Time Domain Constrained (TDC) and Spectral Domain
Constrained (SDC). TDC is used to alleviate the signal distortion when the energy of
the residual noise is below a certain threshold. SDC can be utilized to minimize the
signal distortion under a fixed spectrum of the residual noise. Simulation results
show that our proposed algorithm is able to deal with the arbitrary noise types
effectively.
Index Terms- Karhunen-Loeve transform, TDC, SDC, signal subspace, speech enhancement.
|
|---|