When he doesn’t try to repel the lawsuits from the main labels, Suno is still working on improving her AI music tool. The latest model, Suno V5, is an obvious technical improvement compared to the previous version, V4.5+. But it still seems to avoid a uninteresting emptiness that permeates most of the AI art.
There are some improvements in the field of sound quality, which are undeniable, such as less artifacts and more pronounced separation between instruments. Some songs produced using V4.5+ can combine all melodic parts in a way that the lines between the guitar, bass and synthesizer are at best muddy. But with the V5 mixtures are much cleaner.
During the demo Henry Phipps, the Suno product manager, pointed to a song that we generated a model that contained a synthesizer similar to the flute with what was the effect of ping-strong: “I have never heard it in previous models … stereo. ” Because Suno does not really employ the effects in a time-honored sense, it means that the model defines a specific instrument and introduces the sound of stereo delay, because it was decided that it should sound.
None of the vocals Suno has the edges. Everything is bathed in reverie, layered with harmonies and perfectly on the pitch. Even if you clearly say that not doing these things, the model simply ignores you.
Suno also claims that the V5 understands the species better, although this claim seems doubtful after my tests. With some of my hints, such as “Modern Avant R&B from Glitcha, but funky drums, atmospheric melodic and breathable parts”, nor V5 nor V4.5+ seemed to be a clear winner in providing what I meant (mainly Kelela’s Undress me). They both approached, giving me traces of Downtempo with some moody synthesizers, but they lacked the strangeness I was counting on.
Suno could not come up with what I was looking for with “Early ‘Early’ 90. L-Fi India Rock recorded on a 4-seater cassette recorder with key vocals and slightly with tuning guitars”, but V5 was definitely more independent. Despite everything I tried, I couldn’t force Suno to spit out everything that sounded even remotely like a pavement. Loose noise that I associate with Oblique and enchanted It was not possible to find anywhere. Instead, I got the bombastic “India” rocks with massive riffs and pure power chords. Suno V5 was constantly serving songs that sounded more like arctic monkeys than anything released before the turn of the centuries.
Similarly, during my tests, V5 sometimes seemed to be struggling with specific hints for the era or decade. When I asked for “Krautrock at the end of the 1970s”, V4.5+ basically nails him outside the vocals (more about it later). But V5 often provides synthpop and songs from the 80s, which are clearly more newfangled, even if they have part of this classic DNA DNA Krautrock.
I will say that the arrangements created by the Suno V5 model are much more elaborate. Compared to version 44.5+, there are more disposable musical flowers that stop the situation from too repeated and more diverse song structures. Where V4.5+ is usually satisfied with the basic structure of the verse-refree (with a bridge connected to a good measurement), V5 would often have sections before or chorus, many bridges or failures and generally build on a track offering more arc than just separate sections.
From time to time, he also provided engaging results when remixing existing songs. I wondered the song from the EP, which I released a few years ago (which should probably have been stumbled with the copyright filter) and I look, I do not intend to lie, I liked the way it prescribed part of my guitar solo into a repeated synthesizer motif and turned my huge pads into leading Arpeggios.
But what was missing in all the covers of my song, which I asked for the creation is the raw, L-fi nature of the song that I Recorded in my salon at 3am About six years ago. And this is the type of current motif here. While Suno can imitate some of the superficial features of an ancient recording or human performance, such as tape hissing or breaths, it always seems inactic.
Phipps admits that he has not heard that the vocal model has reproduced the unique imperfections of a real human performance. In early news about the V5, Suno advertised “emotionally rich vocals” and “human emotional depth”, but this phrasing is now absent in materials directed to public. Instead, the company decided to describe the vocal as “natural, authentic”, combining a change into “stylistic choice”.
But even that seems to be an episode. While yes, compared to version 4.5+, the vocals feel more human, they are still stiff. Phipps explained that “when we perceive vocals from Suno [v4.5] To be emotionally flat, I think it is because he lacks certain details that give it this advantage “and that the higher loyalty of the V5 model provides these details.
It is challenging to argue with the technical aspects of this claim – vocal performances are more detailed – but they are still painfully general. Each rock vocal sounds like Imagine Dragons or Mumford and Sons, every R&B song like Dormant Adele or without the charm of Ariana Grande.
None of the vocals Suno has the edges. Everything is bathed in reverie, layered with harmonies and perfectly on the pitch. Even if you clearly say that not doing these things, the model simply ignores you. I asked V5 for “unprocessed emotional solo, and Cappella a female vocal performance without reverberation, without harmony, no effects, just a dry vocal.” Two songs he provided were bathed in reverie, included additional vocalists harmonizing with the first, and one even had what sounded like a bass accompaniment. (Although it may be a voice approaching the bass.) But Phipps was not surprised. “Models do not yet understand the descriptions of specific effects and recording techniques. The most affect the way the vocals affect the lyrics and the overall mood,” he said.
So I fed Suno lyrics that were Just “Rolling Stone” is quite differentGimme Shelter“ To avoid marking copyright violation. In the first place it seemed that he had all the elements that make the original so destructive. A powerful singer shouting a full, blues arrangement, but had all the emotional influence of a dentistry textbook.
When I listen to “Gimme Shelter,“ This is the way Mary Clayton’s Voice cracks When he throws out “rape and murder” during the bridge, which causes me to get it. This completely unused Warble Robert Smith transfers desperation in “Why can’t I be you“And tangible exhaustion in Kurt Cobain’s breath before he provides the last line in”Where you slept last night“This says that he is a man struggling with real demons.
In general, the attempt to make Suno sound “bad” – not tuning, harsh, beyond the key, messy – was in vain. Throughout the whole conversation of the company about how the “natural” vocal of the new model sound, there is a lack of imperfections, which often carry the emotional importance of the performance. The virtual singer Suno still sounds detached. The V5 model may understand that a specific lyric should be sad, but there is no actual emotional relationship with words, because it is a pile of code, not an artist.