On 4/27/24 5:42 AM, Fons Adriaensen wrote:
On Fri, Apr 26, 2024 at 12:34:03PM -0400, Tim wrote:
Would you have any insight into how this product achieves this and the
techniques used?
It must be a combination of a lot of different things, carefully
tuned for the best results.
Their patent application provides some information on what is
likely going on.
It seems to be based on comparing the input to stored
waveforms, and combinations thereof to detect chords.
In the marketing blurb they call this 'AI', but that seems
to be a little bit over the top. Having a reference waveform
for each of the 6 * 22 single notes that can be played on a
guitar can't really be called 'training' in the AI sense.
It's more like what would be called an 'expert system', the
procedures used seem to be explicit instead of being the
opaque result of 'training'.
Then there must separate algorithms to detect pitch bending,
glides, etc.
It's certainly not simple, and an considerable achievement.
As to latency, it's reported to be quite low, but no hard
figures seem to be available. Nor of course has any of the
'reviewers' ever attempted to measure it.
Re. playgin KE's parts on guitar: that may be possible
for some of the monophonic synth lines, but I don't think
you could ever play the wonderful piano parts from e.g.
'Take a pebble' or 'Trilogy' on a guitar...
Ciao,
Thanks very much, good stuff!
Here's Trilogy.
It's a combination of amazing two-handed tapping and MidiGuitar in places.
The product is polyphonic, yes, but I agree he probably couldn't play
complex piano pieces with the piano setting, as good as it is, although
I think I have seen them do a bit of it here and there.
https://www.youtube.com/watch?v=zZwgpFX4uwQ
Tim.
_______________________________________________
Linux-audio-dev mailing list -- linux-audio-dev@lists.linuxaudio.org
To unsubscribe send an email to linux-audio-dev-le...@lists.linuxaudio.org
