Resynthesize any audio file lossily as a sum of sine waves, additive synthesizer style.
- Because it's cool
- Because most generative audio models currently operate on the mel-spectrogram of the source audio. This requires a lot of memory, much of which is probably unnecessary for recreating harmonic sounds (e.g., most musical instruments.) In principle, you can reduce the memory requirements by finding the harmonic peaks in the spectrogram above a given amplitude threshold and modeling the source audio as a sum of sine waves at those frequencies.
Currently, only static spectral profiles are supported. But it shouldn't be a huge lift to support dynamic spectral profiles as the audio_to_sine_waves() function already returns time values.