This study presents a method for separating periodic and aperiodic components embedded in speech signals. The fundamental frequency is first estimated from a frequency based technique using a whitened cumulative periodogram. A simple partial detector is used to avoid octave errors. The pitch detection is robust with respect to high level of colored noise. The periodic component is then estimated via the projection of the signal on the subspace spanned by the harmonics. The aperiodic component is obtained by subtracting the periodic component to the analyzed signal. Numerical validations on synthetic signals show that the presented method successfully separate the periodic and aperiodic components of simulated voice segments, even in very complicated case, such as voiced fricatives, which exhibit low and frequency-dependent harmonics-to-noise ratio. Applications on real speech signals highlight the interest of the technique to quantitatively estimate speech features such as harmonics-to-noise ratio, or voicing degree, as a function of time. Robust tonal and noise separation in presence of colored noise, and application to voiced fricatives