Detection of multiple sinusoids in unknown colored noise using truncated cepstrum thresholding and local signal-to-noise-ratio

Although cepstrum-based harmonics-to-noise-ratio techniques have been widely used in speech processing, they are as yet not applied to detect multiple sinusoids in unknown colored noise. By studying the impact of sinusoids on cepstral coefficients in theory, a truncated cepstrum thresholding technique is proposed to estimate the unknown colored noise power spectral density, which can be used to estimate the local signal-to-noise-ratio (LSNR) in the frequency domain. This paper suggests the use of the LSNR as a test statistic to detect the sinusoids in unknown colored noise. Numerical simulation results show that the proposed test statistic is much better than the existing test statistics in unknown colored noise environments.