抄録
The fixed dimension modified sinusoidal model (FDMSM) was recently proposed as an attractive candidate for compact representation of audio signals in adverse conditions. This paper aims to study the capability of the FDMSM signal representation for analysis and synthesis of speech mixtures as well as noisy audio signals corrupted by highly colored noise of babble and harmonic. Extensive simulation results verified that the FDMSM provides high perceptual quality of the synthesized output signal compared with the conventional harmonic plus noise model (HNM) for both speech mixtures as well as audio signals corrupted by various types of noise.
