Hybrid models of speech have received increasing interest from the speech processing community. Splitting the speech signal into a periodic and a non-periodic part increases the quality of prosodic modifications necessary in concatenative speech synthesis systems. This paper focuses on the decomposition of the speech signal into a periodic and a non-periodic part based on a Harmonic plus Noise Model, HNM; three versions of HNM are discussed with respect to their effectiveness in decomposing the speech signal into a periodic and a non-periodic part. While the harmonic part is modeled explicitely, the non-periodic part (or noise part) is obtained by subtracting in the time domain the harmonic part from the original speech signal. Three versions of HNM are discussed. The objective of the discussion is to determine which of these versions could be useful for prosodic modifications and synthesis of speech.