Pitch Extraction Using Fourth-Root Spectrum in Noisy Speech

抄録

In this paper, we present the use of the fourth-root spectrum instead of the log spectrum for pitch extraction in noisy environments. To obtain clear harmonics, lifter and clipping operations are performed. When the resulting spectrum is transformed into the time domain by the discrete Fourier transform, pitch detection is robust against narrow-band noise. When the same spectrum is amplified by power calculation and transformed into the time domain, pitch detection is robust against wide-band noise. These properties are investigated through exhaustive experiments in various noises. The required computational time is also studied.