The method of compiling speech reconstruction with its modifications which differs in favour way in its good articulation, natural properties of speech synthesized as well hardware implementation simplicity is proposed. The method is a combination of direct and parametric speech encoding and based on spectral-time approach when time intervals between signal zero crossings are used as spectral expansion ones and spectral expansion coefficients - as spectral parameters.

Keywords: compiling speech reconstruction, spectral expansion in functional series, speech articulation and quality estimates.