pouët.net

Youtube link here.. https://www.youtube.com/watch?v=DFteV6YE7F0

The basics of this encoding method have been explained previously. Below is a summary as well as some information on the additional improvements in the encoder and decoder.

The Frodigi method is simply conversion of audio into sid waveforms and frequencies. In very much the same way as wavetable changes in SID music. The two main advantages in this method is the very small file size in the output data as well as minimal cpu usage (the decoder only needs to write a few values to sid registers once per frame

This updated encoder and decoder now supports seperate amplitude per channel (this greatly increases the definition of the audio) I did actually incorporate this method in my older encoder previously but had discarded it due to the annoying clicking that was produced.

A few days ago, was talking to agemixer and he had mentioned that changing the sustain and setting gate off/on produces peaks of amplitude on every change (This was already known previously) However setting sustain value of $f brings down the amplitude level to zero (and then the sustain value can be placed (0-e)

Depending on the current sustain level, Lets say for example it was $04. (and previous value to this was $03), The sustain can be changed to $0f, and then $04. This results in amplitude previously being $03, then falling to $00 and then $04. Using the previous implementation, it would be amplitude $03, then rising to $0f and then $04 (which is more noticable)

Ok that is the decoder out of the way. It sets gate off/on only if sustain is rising and if it is, then there is one of two ways to adjust the sustain.


There has been around a dozen changes and updates to the encoder as follows

Sustain static (keeps current sustain level if possible) and not a specific threshold higher. e.g if current sustain level for channel 1 was $04, the encoder would make sure it would only use sustains $00-$04 or $06-$0e (and not 1 or more above) This reduces definition, but decreases the occurance of sustain fluctuations

Sub-band preprocessing and encode
The low frequencies of the source are simplified and encoded seperately into the first channel
The remaining two channels are encoded together using the second frequency band to recreate the audio

Multiple window overlap. To keep transients sharp and bass smooth, various size windowing options are used. In this example 50% windowing on low frequencies and no overlap in the rest.

frequency refine
One an "optimum" seed is found, the local area is searched in order to lock on further to a more accurate volume/frequency for that particular channel

The basic idea of the encoder is to start of with a population of data and to feed it into the virtual ear. the ear then gives the feedback to the encoder and the population is analysed, mutated, replaced, merged with multiple iterations.

Multiple attempts as well as combining the above methods with stepcount hill climb allow local minima problems to be minimised.

The c64 decoder used in this demo uses 7 bytes of data per frame updating at 50 frames per second (This equates to 350 bytes per second) I am however also using inline depacking in the demo that decompresses chunks of data while the audio is playing (This was to give me some space to put some gfx in)

Will there be a frodigi4? Yes, I have idea's on improving the quality further (In particular smoother) but this would remove most of the naturality further (even though it sounds rather unnatural as it is)