Somewhat busy

Hi folks!

We had several online and offline meetings, apart from our usual working meeting, brainstorming and the rest. I have traveled from autumn to winter to summer and so on, as we are all scattered pretty much over the planet.

Naturally, you are interested who “we” are. One of the names is plainly visible in references, that’s Konstantin Boyandin, our primary workhorse, talking of producing audio content.

Then, there are other developers, including those whose voices we use in the process of songs and instrumentals creation. Photos and so on will follow, but first things first.

Generative content marking

Our entire team supports the idea to label the generative content we offer you in accordance with DDEX recommendations.

We are “pro” transparency. The listener should know what they are listening to, so here I sum up our base principles, in the below preformatted section.

We are heavily using AI tools to produce final audio content (be it songs with vocals or instrumental tracks).

The songs we generate are derived from human-written (translated etc.) lyrics. I.e., our original works or, in rare cases, public domain poems.

The so called Personas (Suno term for spectral features/pitch/manner/vocabulary etc. of a template used to produce vocals) are derived from real human voices recorded (our team's voices).

When editing and mastering are required, our typical tool is Audacity and several other good old tools of trade (also human touch).

We do our best to ensure quality of the resulting tracks, over quantity.

With the above in mind, you might estimate the sheer amount of work to introduce as automated labeling as possible, so that every media file we release would be properly categorized and marked.

Thus our pause in communication with the outer world. Stay tuned, folks, and thanks for your support.

Since the middle of September, 2025, Suno began rolling out the next generation of their transformer model, v5.0.

I have accordingly updated the references and comments in Suno reference repository, check when necessary.

The first impression: the model produces richer sound and allows finer control over remastering. One can try using v5.0 for hard cases such as BGM, dance music based upon genre fusions and the like.

The Shrooms (Rap Story), opus 2567 track is an example of straightforward usage of the model v5.0 to improve a track through covering.

In the previous post I mentioned vocal hallucinations. Out of the four most prominent audio tracks generation services, Suno and Producer.ai (a.k.a. Riffusion) are the best playgrounds to test the vocal hallucination in vivo.

Assuming you can try that in Suno, here’s the definition of a track I have under a name “Waltz of Wires” in my collection. That was the first definition that was capable of producing huge amount of “rich vocal hallucinations” (meaning, very quality tracks sung in “unknown language”).

So, here is definition.

Style of Music:

Minimal techno meets circus waltz, with pulsing beats, eerie calliope, and vocal glitches, female ghost vocals, vocal hallucinations.

Lyrics:

[genre: minimal-techno, circus-waltz]
[style: mechanical, eerie, playful]
[mood: surreal, hypnotic, uneasy]
[tempo: steady 4/4 with waltz overlay]
[instruments: drum machine, synth bass, calliope organ, tuba]
[compression: light]
[vocals: mechanical, glitched, fragmented, female ghost vocals, vocal hallucinations]
[length: 210]
[intro: Clock-like percussion under warped calliope organ chords.]
[structure: intro, theme A, bridge, theme B, outro]
[theme A: Pulsing synth bass with sparse drum machine, overlayed with a playful organ waltz.]
[rhythm: 4/4 pulse clashing with 3/4 organ swing]
[sfx: distant fairground crowd, murmured announcements]
[bridge: Percussion strips away; organ bends into atonal drones, with vocal glitches.]
[effects: granular synthesis, reversed whispers]
[theme B: The techno beat returns, now robotic, while the waltz organ spirals into madness.]
[sfx: distorted laughter, vocal stutters]
[outro: A final organ flourish cuts into silence, with mechanical vocal echoes.]
[fade: abrupt cut, tape reel spin-down]
[end]

Use models v3.5 or v4.0. After you make a vocal hallucination, or just a nicely sounding track, make a persona out of it and continue generating with the same track definition.

In my case, the above approach allows generating tracks where 85-90% of the output will be vocal hallucinations.

The above stops working in v4.5 and higher, but don’t worry: there are other templates (definitions) begetting vocal hallucinations in modern Suno models with high probability.

The next obvious question is “why could one need vocal hallucinations at all?”

I will begin offering good answers for that in my next post. Also, I would like to remind that I keep most useful information on generating tracks in Suno/Producer.ai in this repository.

Hallen is a hallucinated English.

Vocal hallucinations in generative audio are well known. When talking of songs, that means vocals generated unintentionally, usually sounding like gibberish or a “song in unknown language”.

When it’s human-made, it’s usually called novelty song. If you ever listened to, say, “Orbis Mundi” albums, starting with “Adia”, you would find quite a lot of such songs - they sound in something very similar to (language name here), but still sound weird and can’t be understood.

When it’s a work of algorithms, it has no definite name, save the umbrella term “vocal hallucinations”. In my experience, the Hallen songs can be quite an interesting type of tracks. Although the language itself isn’t real and can’t be understood in semantic meaning, the voice still carries emotions. When that blends with a music, it can result in quite an interesting audio experience.

When we started our first attempts of generative music composition (clumsy, weird and awful; there was no Suno or other services at that time), Hallen tracks were generated in old school manner - talking of scatting. Now that the technologies allow us to speed up the process, generating Hallen tracks intentionally is a piece of cake, after one finds proper templates.

What kind of templates? That I will explain in one of the next posts.

Here we are, another lullaby:

Moonlight draps on the bairn’s saft cheek,
Toy-box tune frae the gloamin’ sleek.
Stars keek doon wi’ their shimmer braw,
Haudin’ dreams whaur the nightwinds ca’.

Rest ye still in the moonlit cradle,
Lanterns burn sae mild, sae fain.
Sisters sing at the shadow’s ladle,
Keepin’ watch till the dawn again.

Phonk-beat low as the heart tae sleep,
Echo’d vows frae the deid ones keep.
Silver thread in the cradle’s seam,
Bindin’ close to a guidnight dream.

Rest ye still in the moonlit cradle,
Lanterns burn sae mild, sae fain.
Sisters sing at the shadow’s ladle,
Keepin’ watch till the dawn again.

O, hush, hush, bairn sae wee,
Moonlit cradle carries ye.

I was already asked, if we will re-post all the tracks from the Topic. Some of them, for sure. But not all of them.

But not all of them. Instead, we prepare to begin posting under-the-hood details of how we choose, invent, work on and finally release the tracks. Stay tuned.

The next track we plan to post isn’t from the Topic; it’s kind of by-product, which was good enough to show it to general public.

The tracks we produce are, to some extent, AI-processed or AI-generated (we often use existing, manually crafted - say, via MIDI editor - underlying theme, processed later by a generative service).

Tools

To begin with, the below are links to the services and tools we used at least once to consider. The detailed overview will be given later, as well as the criteria of what and why we use.

So, the list:

Music generators:

• Soundful, paid-for
• Suno, paid-for
• Riffusion/Producer, paid-for
• Sonauto, free
• Udio, free

Spoiler: we only use three of them: one - heavily, one - occasionally, one - rarely. Which is which isn’t too hard to guess.

Mastering:

• eMastered, paid-for
• AI Mastering, free

However, we plan to share our knowledge and experience by adding to and elaborating the docs of Suno reference.

So, before we post a new track analysis, feel free to share your thoughts about the published above reference. It took quite a time an efforts to compile it.

As promised, I provide detailed explanation of the lyrics for “Midnight Waltz” track (Wilds Vol.1 collection), which has been uploaded to YouTube.

To begin with, the lyrics in full.

1. At Braeside Kirk, the fog cam’ doon,
And stitched the sky wi’ silver seam,
She sang beneath the hollow moon —
A spell to keep the bairn in dream.

Chorus:
Lie still, lie still, sae lang the nicht,
The lassies guard ye, pale and mild.
A whisper’s lantern, faint o’ light —
Shall lead ye back, my slumbering child.

2. The yew tree bends tae hear her song,
A croon sae low it stirs the clay.
Her sisters hum where shades belong—
And cradle stars that fell away.

Chorus.

The song is in Scots (even though it doesn’t prevent most of us from understanding it at once, even if we only read and speak English.

It looks like a lullaby, weird or not, out of thousands of them. But let’s delve into the meaning, line by line.

Verse 1

At Braeside Kirk, the fog cam’ doon,

Braeside Kirk: A rural church at the foot of a hill (brae). In Scots ghost lore, kirks with graveyards are liminal spaces where the living and dead may meet.
Fog: Often seen as a veil between worlds, allowing spirits or fae to cross over.

And stitched the sky wi’ silver seam,

Silver seam: Alludes to protective silver thread charms in Lowland superstition — a sewn “seam” sealing the night sky from harm.

She sang beneath the hollow moon —

Hollow moon: Waning or crescent moon, potent for charms, and sometimes considered a portal to the Otherworld.

A spell to keep the bairn in dream.

Bairn: Scots for child. Protective lullabies (often charms) were sung to prevent a child’s soul from being “taken” by fae or wandering spirits during sleep.

Chorus
Lie still, lie still, sae lang the nicht,

Sae lang the nicht: Traditional lullaby phrasing (sleep until day), reinforcing protective enchantment.

The lassies guard ye, pale and mild.

The lassies: Here likely supernatural female guardians — echoing faerie women or bean sìth who protect sleepers.

A whisper’s lantern, faint o’ light —

Lantern: Protective guiding light; “whisper’s lantern” evokes a spirit-light, guiding the dreamer’s soul home.

Shall lead ye back, my slumbering child.

Lead ye back: Ties to belief in dream-soul journeys — the idea that a sleeping person’s spirit could wander and must be safely returned.

Verse 2
The yew tree bends tae hear her song,

Yew tree: Sacred in Celtic and Scots tradition, planted in kirkyards as guardians of the dead; symbol of immortality.

A croon sae low it stirs the clay.

Clay: Grave-earth; the song’s power reaches the dead below. Crooning links to lullabies and funeral dirges (caoineadh).

Her sisters hum where shades belong—

Shades: Spirits of the dead; “sisters” could be other guardian spirits or a spectral choir.

And cradle stars that fell away.

Stars: In poetic Scots lore, stars can represent souls; “fallen stars” are lives lost, “cradling” them is sheltering the departed.

The subtle language of known images, especially if they are known to us since the early childhood, can help us to reach every listener much deeper and more thoroughly, leave a touch of something long known.

Do you use such subtle language? Well, OK - did you ever study your own lyrics thoroughly, to see more than rds can directly reveal?

You might notice quite a number of tracks published during the first three weeks of my online presence on Spotify and the other services.

That was intentional. “Seeding” helped me to understand what tracks have a chance to find their listeners quicker.

Before commencing the publishing, I have composed and prepared for publication quite a number of tracks. Now I switch to much slower delivery schedule. It all depends, of course: if some bright unusual ideas visit me, I won’t oppose the urge to publish something out of schedule.

Stay tuned!