Home Creators Posts Import Register Favorites Logout

Content

It's my second month that I 've been using SUNO and I am still impressed with the results as Suno lets me create more or less the music I always wanted to hear and which I was not able to create on my own due to lack of equipment, time and especially lack of talent. The output can be sometimes pretty random though. You tell Suno not to use this or that instrument and in the next creation it will exactly use that one as if it would like to pick on you. Also one day you can have stunning results with particular prompts, while the other day it only puts out junk and after my first euphoria I meanwhile noticed lots of flaws which makes the output close to unusable if you really intend to use it for professional sounding music.

I wanted to share this publicly as I think it's quite an important factor regarding Suno's music creation algorithm for those interested using it, also because a premier plan can become quite expensive over time and wanted to demonstrate a severe and mostly consistent issue which prevents Suno being a serious tool for ambitious music creation.

The music virtuosity of Suno is still stunning as it can output complex chord progressions with ease that still make sense and the arrangement capability is impressive as well, it is also really capable of combining different genres or styles and putting them together into something new. In many cases it would be hard to tell for me if some song is AI generated or "real", but you will always notice because where it is lacking is the basic sound quality. This usually could be fixed with some post processing, to some degree at least, especially if you can work with single stems and edit those separately which in my case would be overkill, because the music creation algorithm from Suno is flawed to begin with and cannot produce neither tracks nor single stems with a constant quality. Already after half a minute of a song you notice that the quality becomes worse and worse. After 3-4 minutes most songs sound chopped like through some low-bitrate distortion and other side effects, some tracks lose their stereo width, become much bassier towards the end and shift their frequency spectrum compared to the beginning of a song. If you manage to make the beginning of a song sound great, it will definitely sound off towards the end. With many of the created songs I noticed some kind of comb-filtering which is similar to the effect if you play the same wave file on top of each other with a slight delay and modulate that slowly. The effect from Suno is very similar with a creeping shift in frequencies the longer the song will last.

I am not sure if this is some glitch or bug, but I rather think the entire music creation process is so complex, that the algorithm simply cannot keep all those factors well under control.

I wanted to show the issue with some example of a rather acoustic track which I had planned as series for a relaxing acoustic smooth jazz collection. First you hear the raw output from Suno completely unedited.

Next is an edited version where I tried to make it sound more natural, and although the beginning of the song sounds became much more usable, you should notice that already after 10 seconds the entire characteristics changes and gets worse and worse the longer the song lasts. Towards the very end it already sounds pretty distorted also because it got much bassier, there are lots of digital artefacts and other digital mess which are impossible to handle.

I also noticed that issue with others' songs, especially those that are supposed to have a higher sound quality to begin with, also those start getting worse and worse sounding over time with exactly the same issues I am having.

I was really hoping to be able to use Suno for some serios music creation, but this sound quality issue is pretty hard to work around.

I would like to share another example with you of a song which I used for my Robot album:

Here the raw Suno track:

And below my edited version with an additional intro which was pretty hard to to make it sound usable at all as the upper mids and treble are so messy in the original track, that I had spent some hours to achieve a more natural rendering, unfortunately the sound changes bit by bit towards the end, therefore I mastered 3 different version which were optimised for particular sections of this song and then faded between these 3 sections to achieve a single track. It's far from perfect, but aside of the sound quality issues, I am impressed with the output, what do you think?

So far I have reached the limits of what Suno is capable of for my own purposes and mostly this quality shifting bug is what prevents me of investing more time into it, because I can never achieve the sound quality I am really aiming for. If a song cannot keep the same quality throughout its length, all this becomes just a joke. With many songs this issue is not that obvious but I get it with all songs to some degree.

Files

Comments

Josh Barrie

What do I think? Well, I immediately think, "Your call is important to us... Please continue to hold, and your call will be answered in the order it was received..." 😂

Domingo Martínez

Hi Oluv First of all, thank you for your comments on a previous post about my SoundCloud track created with Suno. I agree with everything you said about creating music with Suno, but despite everything, the level of fun it brings me is immense. I almost always use it by uploading progressions of 4 or more chords and using the beta 'sample' option, and in many cases, it generates truly wonderful tracks with an absolutely stunning level of arrangement and production. If they ever fix the problem of the progressive sound degradation, it'll be game over ;)

Domingo Martínez

And about your song by Suno: simply stunning and with a considerable improvement, and I think enough, over the original sound

oluvsgadgets

I also think that "traditional" musicians will have some issue soon. I mean Suno is able to create music that I enjoy more than most commercial stuff released daily. I hardly listen to Apple Music anymore...

Javier

The main issue with Suno is that the AI generates the entire track as a single, baked audio file. Instead, it should generate individual instrument stems and output them in a MIDI format. If it worked that way, the audio quality would be much better, and the songs would sound far more structurally consistent.

oluvsgadgets

You can download single stems or midi files too.

Javier

Yes, but It doesn't produce like we humans do, It generates a whole baked audio file instead of composing It like we humans do, It's not the same. Anyways I'm surprised with the new v5.5, listening to my voice sing was quite incredible and the quality has improved a lot, you should do some testing now.

oluvsgadgets

I did but it still sucks with deterioration towards the end I even made my own model and this didn’t change anything. Also didn’t manage to upload my own voice as it couldn’t verify it, I recorded my voice with a professional mic while the verification goes live through the PC mic and maybe it cannot perform the verification properly.