SpectMorph is a software which analyzes wav files and builds a frame based model of these wav files, where each frame contains information about the spectrum. Each frame is represented as a sum of sine waves, and a noise component. There are command line tools like smenc and smplay for encoding and decoding SpectMorph models, which are documented in the manual pages. Technically, these tools are frontends to the C++ classes in libspectmorph, which are documented here.
The encoder is implemented in SpectMorph::Encoder. It can be used to encode an audio file; the frames (short snippets of the audio file, maybe 40 ms or so) are then available as vector containing SpectMorph::AudioBlock objects.
Many SpectMorph::AudioBlock objects are needed to represent a whole sound file, and the SpectMorph::Audio class is used to store all parameters of an encoded file, along with the actual frames. Information like the sampling rate (mix_freq) or the original note (fundamental freq) are stored in the SpectMorph::Audio class. Functions for storing and loading SpectMorph::Audio objects exist in namespace SpectMorph::AudioFile.
The decoding process is frame oriented. For each SpectMorph::AudioBlock, a SpectMorph::Frame object needs to be created. For each frame, a SpectMorph::SineDecoder and a SpectMorph::NoiseDecoder can be used to reconstruct (something which sounds like) the original signal.