I wanted to experiment by including a song of mine on Suno, featuring a woman’s voice in Arabic and Italian (Radiodervish style). It was a game, but I really liked the voice. The lyrics are mine, but I don’t want to politicize this wonderful forum. Whatever you think, happy music.
Daft question - is it possible to provide the song and let Suno provide versions of vocals on top of it? Or how does this work?
Hi, that’s exactly what I did. I gave Suno my finished song and (importantly) asked him to have an Arabic female voice sing the lyrics I wrote, the first part in Arabic, and the chorus in Italian. In the end, I noticed that Suno changed just one chord in the chorus. I’m sure he’ll improve in the future. Already, the fact that I can have the singers sing in Arabic and Italian, and not necessarily in English, is an excellent result.
I’m clarifying because I haven’t done it before. I asked Suno to do the female vocals, and he also added a slightly “untz untz” backing track, which I removed with Ultimate Vocal Remover. I only kept the vocals and inserted them into my song. So the music and lyrics are neither AI-generated nor rearranged. The music is mine, and so are the lyrics.
Got you. Nice. Sounds like a great tool to use for adding vocals to instrumental tracks.
There is a fair amount of scepticism on the ethics of Suno due to its failure to credit it’s source material.
From what I understand from the Machiavellian instructions, strictly in English, the songs created remain available to anyone. Honestly, this doesn’t worry me since what Suno creates has nothing to do with my original song. Suno tends to create music in line with trends… hence sequencers and UNTZ UNTZ, which I don’t love. In the case of this song of mine, I only had to insert a Dmaj into the chorus, which wasn’t intended but honestly fits perfectly.
NEW VERSION!
My standpoint, on the currently disproportionately hyped subject of AI-generated audio content, is that the only sectors actually impacted will be (they already are for what matters) incidental music, low-tier cinema soundtracks, videogame OSTs, advertising tracks and mainstream pop.
Unfortunately, these areas of culturally lesser (or neutral) output are also the ballparks of everyday professionals, who struggle to earn a living out of ordinary musical engagements.
For a fair number of technical and aesthetic reasons, which discussing in detail would transcend the scope of a post, art composition is - and presumably will long be - beyond the grasp of generative LLMs, except maybe in the foreseeable future as mere surface-level mimetic emulations of existing works.
What music AIs seem only to do, for the moment, is cheesy jumbles of scanned audio materials, available on the web or on dedicated example repositories. The complex self-imposed semantic rules of connection and deflexion, characteristic of a human artistic brain, elude completely the inherent functionality of generative neural engines, for the time being and arguably for a long time to come.
Hi @Aethermind, I was eagerly awaiting your comment, which I agree with. I must say, however, that for a song like mine—harmonically simple, but which I wanted to make “passionate” with a female voice singing in Arabic and Italian—I was quite happy with the result. Obviously, I eliminated the music the AI proposed using “ultimate vocal remover”… it was scandalous… unz unz and sequencer and it sounds really ugly. But for me, who doesn’t have an Arabic singer available (if I still lived in Bologna, perhaps I could have had one…), this solution was a great help. I had to insist with the AI to have an “a cappella” voice, which still included some instruments, and to make it follow the chords that made up my song. So on this song, the AI is ONLY and EXCLUSIVELY the female singer. The parts of voice without lyrics are a VST from Native Instruments (Voices of Sahara).
I think @riban meant the thousands of unheard and forgotten voices that were used to train the model creating the vocals.
I had the opportunity to try a VST that combines vowels and consonants to create a word. Naturally, it’s only in English. You can tell the sentence is a little fragmented, but the result, for those who don’t have access to a singer, is enough to carry out an amateur project. The limit was the language (sorry, English speakers). I tried inserting syllables to create lyrics in Italian… and the result was terrible. Suno does this job much better in other languages. Is it right? Is it wrong? I can’t say, but it allowed me to create a song in Arabic and Italian that was decent for my tastes, allowing me to focus my efforts solely on the music and not on searching for a native Arabic-speaking singer who also speaks Italian. Of course, the artistic integrity of creating something with our own efforts will remain a primary issue, but every sampler imitates a synthesizer, a bass, a guitar, so even these sound “fake.” It all depends on how you use them. I love bands that actually play, but I’m alone here and I need to find help from a virtual singer… and today the message was more important to me than perfection (I shouldn’t have written this, sorry).
I also add that this song is licensed under the Creative Commons Attribution-NonCommercial 4.0 International — CC BY-NC 4.0 license, so it is completely free and I will not earn a penny (to stay on English) for this work.
I have several regularly paid sound banks featuring female artists’ vocals. For example, “EW Voices of Soul,” recorded by the famous singer C.C. White, who I assume received compensation for her work, and “Voice of Sahara” by NI (which I use in the final part of my song). So, if we’re talking about royalties, are these sound banks usable? I hope I’ve understood the point better; if not, I apologize, as always.
Sorry, dear @Lanfranco , I don’t think anybody is blaming you for using Suno, most people here are rather blaming Suno for disrespecting uncredited artists used for their training data, their gamification of music approach to make making music a consumer experience and AI in general for being the future death of open source. I also think that most of the criticism about AI in music is not about the “fake” aspect, since you’re completely right that making music or sound in general is often about mimicking.
I don’t think anybody wants you to justify your usage. It’s rather only unlikely that AI usage in creating music will resonate.
I apologize to you @hannesmenzel and the other users for my misunderstanding. I agree with you all. I also think AI is ruining music, and I reiterate the concept of “intellectual honesty” in creating music with our own heads. However, the problem remains of using sounds that we can’t have as standalone instruments. I use the piano a lot, but I don’t have a grand piano; I use the MiniMoog a lot, but I don’t have a real Moog… I’d like to use vocals a lot, but mine sucks, so I have to use sampled ones. I reiterate that what I did in this piece of mine was to use a female voice. Everything else was created by me: the chords, sounds, performance, arrangement, recording, and mixing.
It’s really less about the intellectual honesty in creating the song but more in how the AI was trained. I’m confident that sampled instruments are another category. I feel you want to strengthen the point that your song using AI vocals is an original composition anyway, and I think this is true.
When talking ready made vocal sample libraries we’re not talking about intellectual property anymore, but that’s the point where I’d stop for the honesty issue (although it was my first experience in computer music when someone introduced me to Magix Music Maker and it’s ready made sample loops).
Lack of compensation anyhow has a long history. I think the drummer unintentionally created the so called “amen break” died poor, and the woman which became the voice of Siri didn’t even know what happened when she was hired for vocal recordings for a couple of bucks.
That someone profits from the work of others is part of human history. Unfortunately.
Hi @Lanfranco, thanks for noticing my contribution to the subject. In a nutshell, I am more concerned by the (potential) reduction of average music writing to a consumer-grade Internet platform experience - as aptly stressed by @hannesmenzel - than by the slightly obtrusive topic of AI sources ethics.
Again, I honestly don’t think that an LLM engine will be capable of writing something on the same degree of compositional prowess as Ligeti’s Aventures very soon, but it will be arguably able to churn out scores of Beatles-like 1960s beat tunes, although certainly not with the same grace and at the same artistic level as the famous Liverpool quartet.
As for the tendency of mid-tier and “practical” artists to plagiarism, this is not inherently a problem of AI age and its massive pre-training requirements, since it’s being aggravating the music industry scene for centuries. What matters, in order to learn an aesthetic language, is to detect the mid-depth syntactic constructs that give shape to the perceptual surface upwards, but are controlled downwards by deep structures of meaning, which affect the comparison and choice of alternative grammatical solutions. Such a kind of contextual and intellectual awareness is simply beyond the reach of generative AI, as it stands now.
Cheers 
I really like, @Aethermind , the way you approach, with elaborate syntax and great diplomacy, a question that has plagued us for many years… popular music today sucks. And that’s why AI can afford to create artistic “masterpieces” in line with the latest hits. This is my humble opinion, born from the mind of a poor musician who is aware of his limitations. Almost all of today’s music has been preparing us for the advent of AI for years with a “Desert of the Tartars”-style simplification that offends even my unprepared ears. This will be an unstoppable drift, like the advent of rock ‘n’ roll after great jazz, and as always, there will be a return to more beautiful music. Everything in the universe is cyclical.
(I can’t imagine what Google Translate has translated of all this… don’t hate me.)