The recent research trend towards the use of harmonies/sinusoid based methods, in order to exploit the fine spectral structure of voiced speech, cannot be questioned. This paper discusses the state of the art in this area, both in terms of analysis-synthesis methods and of their application to coding. The key points are: • - Harmonic modelling is a very efficient tool for voiced regions, producing synthetic speech of very high quality, but being simultaneously prone to pitch and voicing errors. The main disadvantage of harmonic coding is the need for an alternative method for unvoiced regions. ATC is a natural choice. In this paper, an 8 kbit/s simulation is presented, using hard switching between harmonic coding and ATC. • - Sinusoid based modelling extends the basic analysis-synthesis framework to unvoiced and transition regions, by removing the constraint that the sinusoids be harmonically related. When it comes to coding, however, it still has many unsolved problems. As a conclusion, some guidelines for future research are discussed.