pouët.net

Since the release of Frodigi, I had implemented many changes in the encoder for the PC that generates compressed audio data that can be played back on a C64.

For anyone that has not heard the previous frodigi demo, this demonstration produces "audio samples" by changing waveforms and frequency only once per frame in order to recreate the audio.

More technical details to follow. Use new SID for higher quality

Thanks all. Some very brief technical information as below.

Ok some technical information as follows

Current Encoder capabilities as follows

25hz/50hz mode (25hz mode has half the data rate and aimed mainly at speech)
12bit frequency in each channel normalised to $0000-$ffff
usage of all 4 waveforms
optional usage of empty channels for higher amplitude resolution
Greedy or Step counting hill climbing mode with variable parameter
Optional window overlap mode x2/x4 (helps smooth and reduce distortion)
Multi-attempt mode for higher error reduction
Optional brute force volume in conjunction with selection/hill climbing
Single channel mode utilising sustain for amplitude changes
Ability to use only 1 waveform for the three channels/1 channel for higher compression

In very basic terms, the encoder reads an audio chunk equivelant to a 50th of a second and initially feeds a seed of data into the analysis engine which in turn responds and
allows tweaking and improvement of the small data set until it is refined to a sub-optimal set of frequencies and waveforms for the 3 SID channels)

There are many other adjustments that take place in order to improve on quality including multishot attempts and future generation creation as well as variance static.

There are still many improvements to be made which will be released in the Frodigi3 demo.. Yes there will be another one.

The c64 data in this example demo consists of 6 bytes bitpacked data per frame (equating to 300 bytes per second) After final compression, data size is at around
180-190 bytes per second

The decoder merely reads the bit-packed data, normalises the frequencies, extracts the waveform data and volume data for each chunk per frame

Code was put together in half an hour and i am aware that the decoder is not optimised in any way whatsover, but i guess can be optimized to use only a few rasterlines at most per frame.

To reduce clicks due to amplitude changing in old sid, i am interpolating the volume data inbetween frames. Preferably use a new sid and a real c64!