So everybody is touting about AI chord analysing but when I ask about it folks get quiet. I think it simply doesn't exist yet! Typical AI hype lol

 

Samplab is considered AI. Try that one. It's not "typical", it's you want someone to do your research and testing for you. Why waste the time, because it sounds like you want to argue with someone about something you haven't even tried yourself.

 

Cheers, I will check it out, maybe it's better than Melodyne who knows

 

For most users, chord detection using Melodyne is almost an afterthought. You are using a tool you already have around which works (the best), but it's not the main reason to learn Melodyne and certainly not to buy it. You can use Melodyne to even learn chords of a song to play them, but others have already mentioned tools that fit that use case better, like Decoda or others like it. Samplab does a better job extracting polyphonic midi data from an audio file than most conventional audio to midi options, but it is aimed at users who are using samples as if they are crate-digging and then snatching just the notes to edit, modify or re-voice with other synths, drum machines, etc.

So the question is not really which is "the best", but which is the best tool for the user's specific use case. Even monophonic midi extraction can work fine, if you are able to manually recreate the full chords yourself in piano roll or using something like Scaler.

 

Fun fact: I loaded in deCoda Dr. Mabuse previously mentioned in this thread, it gets (to me) the wrong key scale: Gm, it's pretty rare in track with no modulation, beside the intro glide and other minor embellishment.

To my ear is Cm.
I tried to prove myself wrong giving trust to the software, but I couldn't.

So unless proven the opposite, all these tools are pefect as valid aids to trascribe songs, not the replacement of the human knowledge of music theory, though I wished there was such a perfect software.

 

Thanks for the info, I appreciate it. Yes I have actually bought Melodyne and use it for other stuff too but the chord / polyphonic notes detection works surprisingly well, IF you have clean audio. I wonder if these AI tools can work also with stuff that is not so clean for example with other tracks playing at the same time (drums, voices etc etc). Also with Melodyne the resulting midi is not so clean, you have to always edit it but you get the notes out of it most of the time which is most important thing. Basically I am using Melodyne now like you explain it, to extract the polyphonic notes from polyphonic audio but it's not always so good if the audio is not super clean. And it's not always so clean with some juicy vintage tracks. So wonder if AI tools are better at this? I wonder if anybody has compared, I will try it myself for sure soon!

 

Last edited: Jul 31, 2025 at 9:31 AM

There are x different versions of the Propagenda - Dr. Mabuse Track

Original: Cm

The Last Word (Strength To Dream): Gm

 

These were from suno midi stems

 

For a full song, normally I will use RipX to stem separate the material first. Once you have the wav files, you can even clean them up in a wave editor with EQ, or even RX it. Most of the time when I use Melodyne, I will put Antares Auto-Tune before it as an insert FX on the channel strip. It can help with sight variations in the notes' pitches, so they are not detected as new note fragments; and I always do it when reworking a vocal in Melodyne to make less work of it. Autotune shifts things more naturally (when not used in hiphop doses), and your moves in Melodyne will also be smaller. It will result in less artifacts. Think of it like Pitch Compression; instead of a regular compressor being made to work too hard and generating artifacts. Two compressor instances doing less work each. Same idea. By the time you transfer the Audio into Melodyne, it is pitch corrected already and cleaned.

When you convert your stem to audio, sometimes you will still end up with fragments of notes in your midi file. Stick an arpeggiator on the channel at a very fast rate like 1/32, or higher; and usually it will completely ignore the fragments. Bounce it and re-separate it again and it will be as good as you are going to get it.

Samplab seems to ignore all the fragments, making it a nice option with less work to do manually. The more simple the chords in the music, the better job the AI will do. I still consider Melodyne "the gold standard", but it is not perfect either.

 

I listened to The Last Word and right, to my ear is indeed Gm.

However, the version I fed to deCoda was exactly Dr. Mabuse.
Recognized by deCoda as Gm, it didn't fit, it was Cm by ear, and I assure you I can easily get the tonality of a song (and this is a simple song) in few seconds, plus modulations in case are there.

There is a recent thread were the pros and cons about these tools were discussed, but I can't point it at the moment.

BTW I'll try to feed The Last Word to deCoda and see if it gets it right (Gm).