Right now, my implemented Soundfont reverb is pretty simple:
- A sample are read from the delay line at a specified tap (right now 20ms (predelay) and 250ms (which is the maximum tap, the size of the delay buffer))
- Said sample (a reverberation) is added to the raw reverb result (which starts with just the sample from the chorus (or raw instrument rendering with all MIDI effects except reverb) stage output) for adding as a new sample to the delay buffer after all taps are handled. It's also added to the result for rendering (from the previous stages). The final tap also pops a sample from the delay line if it's full, to make it become a proper running delay line.
- After the final channel is handled, the raw reverb value is applied a volume setting (for each reberberation in time), then low-pass filtered (to muffle the reverberation) and written to the delay line's FIFO buffer.

The predelay effectively delays the incoming signal to cause an early reflection on the wet signal.

But it seems to still like some kind of echo only? Not that reverb-y?
Is some extra processing required on the signal fed into the FIFO buffer? Maybe some all-pass filters with chained outputs to inputs? But at what frequency or gain should that be?

The reverb volume for feedback is the reverb Effects Send modulator output (scaled to 0.0-1.0 from 0-1000). The reverb filter is set to 20% of the initial low-pass frequency. It isn't known exactly what's best for this.

Also, my reverb simply stops the signal generation after the note is finished playing (volume ADSR idle, thus volume being zeroed) and the predelay plus 5 taps worth of time has elapsed, no matter what's playing on the reverb at that time. Otherwise, rendering would continue forever, even though it's (probably) silent.

Anyone knows a bit about software reverb and can help? Of course I need to keep it relatively light processing like it's now.