Behind the Fake Artificial Whisper

You might already have guessed that the band Artificial Whisper was synthetic. There's no dusty attic where the musicians met. There's only me, my computer, and my curiosity about the line between human and artificial creativity. This project is an experiment, and this post is me clearing out that attic.

Attack of the mimics

For years, I've been fascinated by the idea of making AI output work that could pass as a human-made product. Back in high school, I tried to create an AI-chat that could carry on a conversation concerning a school day - and failed, mainly due to my missing understanding of transformer architecture. We've now reached a point where AI models are so powerful, their product can be indistinguishable from what a person might have created. How much in this post could be AI-generated and how much are actually from me – the very real, authentic human author?

Lots of the AI-generated content sent out into the market today is slop; low-effort, generic content meant to clog up feeds and steal the likeness of others – and worst of all – your time. With Artificial Whisper, I wanted to see if I could use AI tools with care and taste, to create something intentional, something specific, that I myself could enjoy and share with others.

The complexity of music made it a fascinating challenge. We see AI-generated images and videos everywhere, and there's a healthy conversation concerning labeling synthetic content. But until very recently, AI-generated music still operated from in the shadows. Daniel Ek, Spotify’s CEO, stated in this great video by Cleo Abram that we simply do not know how much of the new music on platforms like Spotify is already AI-generated. However, bands like Velvet Sunset is now pushing the third album in their first month, openly generated by AI.

Fake Realities

Rather than judge, I hope to expand the conversation about how AI-generated music can be used as a tool for storytelling and connection. To get your attention and get you thinking, maybe even evoke a feeling. Why was their only one spectator in that theater? How is it to just disappear into the mountains? What happened to Soren? At its heart, great storytelling is a great lie – I’m telling you something, and that I’m telling you isn't true. However, when we engage with the tale, there's a shared agreement: we, the audience, willingly suspend our disbelief to enter a world that isn't strictly "real." We enjoy the journey and accept the rules of the world. But if I told you a fake story without ever letting you know it was fiction, even if I mimicked reality – well then, I've just lied to you without your consent. That's manipulation.

After the modern rise of AI, it is now easier than ever to create convincing personas with backstories and actual interactions in chats. If audiences aren’t told they’re experiencing a story, is that okay? The answer seems straightforward, yet, human influencers themselves tend to create a persona and tell a story through their platforms without always making that clear, maybe even wearing an expected mask.

The internet is only going to get more synthetic. AI artists, -influencers and other automated content are only going to become more common. I think it's important to show how we can engage with this technology responsibly. To create a believable façade so the story is enjoyable, but thin and transparent enough that the reader knows what they're engaging with - what they are spending their time on.

The band's name, Artificial Whisper, showcases how the commitment to transparent navigation of the "fake" even extended to the name itself. The band's name was originally "Silent Whispers," however, it was changed on a suggestion from my girlfriend to make the AI aspect explicit - a human touch for the AI-band.

Sleeping Through Endless Nights

Working with "A Story Told in Monochrome", I felt this tension. Each song started with a concept from me: a title, a brief narrative sketch, or the core emotion I wanted to convey. From there, I utilized language models to generate lyrics drafts and used generative models to create songs, based on those lyrics and a style description. My effort was to sift through this AI-generated text and songs, combining ideas, adjusting tone, and curating until it resonated with the album’s intended voice and my vision. That blend of narrative folk and cinematic indie was what I was aiming for, greatly inspired by bands like Lord Huron. There is something unearthly about it, making it great for telling the alternative reality story of Artificial Whisper.

To translate my vision into AI instructions, I developed reusable rules for controlling the mood. For the lyrics, all prompts were suffixed with the command: "The song should be poetic, beautiful but eerie, but not on the nose - Words such as 'dread', 'danger' and 'terrifying' should not be used."

I had found without the note on the wording, all eerie songs would end up being about a feeling of dread or containing a velvet color - with those literal words – I wanted a bit more imagery. Similarly, the style prompt for the music was described as "dreamy indie folk, indie rock, Atmospheric, layered instrumentation, rhythmic fluidity, melodic sensibility, blend of Americana, folk, and vintage rock influences, twangy guitar tones and ghostly backing vocals," which was alternated and mixed with a few other wordings to ensure the songs sounded more apart from each other.

Of course, generating music through AI tools wasn't a straightforward path. SUNO.AI generates primarily a whole song, in this case taking the lyrics and style-prompt as input. Sometimes the lyrics about a missing person in the Alps was played to the tunes of a happy guitar celebrating Friday. Other times, a song would emerge 90% perfect but with one jarringly out-of-key section. Even though the tool SUNO can help you surgically “fix” parts of a song, the tool is not perfect, and hitting what you seek exactly is almost impossible, if it was not a part of the original song.

To make it sound like all the songs came from the same band the voice of the singer is essential. I regenerated the same song over and over, utilizing a song I liked as input, to get closer to "the sound of the band" without ending up having songs sounding too similar. These iterations led to unexpected compromises. Sometimes, in this band, it’s the bass player who takes the lead vocal.

Few songs, such as "Interlude" and "Homesick" was picked apart, taking elements from different generations and combining them into a new version, handpicked to the detail and removing unwanted AI artifacts. I want this story to be told in more colors than just monochrome. I want every moment of the listener’s time to be met with intention and care.

A Signal in the Static

I distinctly remember listening to ‘A Story Told in Monochrome’ and, to my surprise, genuinely liking it. That’s why the album took its name from that very first experience. Soon after, I got those AI-melodies stuck in my head, replaying on their own.

While algorithmic catchiness is not the same as genuine creativity, perhaps a "human experience" is reflected in the output, when layers are added through intention and storytelling. I decided what to include, what to change and what to share. And that brings me to the final question of this whole experiment: Is my opinion, my taste, my direction and my production enough to be considered the "human experience" in this mix? I’ll let you decide on that one.

I hope it sparks some curiosity about where the "human" ends and the "artificial" begins—a question that's always been at the heart of art, but now feels more pressing than ever.

Thanks for taking part in this journey.