Beyond “Orchestra” — The Sounds It Cannot ReachFurther Thoughts on Personas, Part I English translation of #65 for Non Japanese Readers

It has been almost six months now, I think, since around Part 46 of this series, where I published a list of instrument personas. Around that same period, I also wrote several articles discussing the concept of personas in general.

That was during the transition from v4.5 to v5.0, and given how rapidly things change in the current AI music landscape, it may already feel like ancient history.

Yet, fortunately or unfortunately, much of what I wrote back then still seems applicable today. Many of those observations continue to have practical value, at least in my experience.

So this time, I would like to revisit the topic of personas once again.
Before we begin, however, I should clarify something.

When I use the word persona in this article, I am not referring to what Suno v5.5 now calls Voices—the AI singers that most users associate with the term.

Of course, for many people, “persona” primarily means an AI vocalist. That feature has become a normal part of the platform, and most users are already familiar with it.

The list I published back in Part 46 was actually about something quite different: instruments.

It was a catalog of instruments that appeared to have been learned by Suno, focusing primarily on classical instruments such as pianos, violins, and various wind instruments.

There was no official documentation available. No published list existed, and the developers have never disclosed such information publicly.

Furthermore, one characteristic of AI-generated audio is that individual influences are often subtle. Differences certainly exist, but they are rarely dramatic, making them difficult to hear with certainty.

For that reason, although I compiled the list carefully and tried to distinguish these differences as objectively as possible, I understand that some readers may have questioned its reliability.

That is perfectly fair.

Still, I have no intention of reopening that debate here.
I devoted a considerable amount of time and effort to that work, and I remain satisfied with what it represented at the time.

What matters for today’s discussion is that the list was only the beginning.
In fact, there was a much larger continuation that I never fully explored.

Although I barely mentioned it in those earlier articles, the concept of personas extends far beyond voices and individual instruments.

The internal mechanisms are not publicly disclosed, so nobody outside the company can know for certain. However, after considering various observations and pieces of information, I have come to think of personas as something like traces of learned acoustic identities preserved within the model.

AI singers are perhaps the most obvious example.
Their voices appear to be derived from large numbers of recorded performances that have been processed, blended, and abstracted during training.

Some of those recordings likely came from professional singers, while others may have originated from amateur performers.

At the same time, it also seems likely that safeguards have been implemented to prevent direct reproduction of specific individuals.
Personally, I suspect that various forms of blending or transformation are involved, though that is only speculation on my part.

What we do know is that attempting to prompt a specific singer’s name in order to reproduce their voice now typically triggers warnings or restrictions.
From a historical perspective, this appears to have become stricter around the time copyright-related concerns began receiving serious attention.

But that is a separate topic.

Since AI singers are already a familiar and established part of Suno, I will set them aside and move on to the real subject of this article:

orchestras and performing ensembles.

This raises an interesting question.
Why are orchestra names generally usable when individual singer names often are not?

To be honest, I am not entirely sure.

One could argue that orchestras and ensembles possess a kind of collective identity. Many have long histories, established reputations, and organizations capable of defending their interests.

In principle, that might make them just as sensitive as other artistic entities.
My own speculation is that several factors are involved.

Classical music remains a relatively niche field compared to mainstream popular music, and within classical music itself, orchestras often receive less public attention than composers or famous soloists.

Whatever the reason, the practical reality today is that most major orchestras and ensembles can still be referenced in prompts without difficulty.
There are exceptions, of course, but generally speaking this remains possible.

Before moving further, I would like to ask readers to think about something.
How often do you use the word “Orchestra” in your prompts?
Perhaps you have written phrases such as:

• Orchestra

• Epic Orchestra

• Pops Orchestra

• Symphony Orchestra

• Cinematic Orchestra

and countless variations thereof.
These terms are used everywhere.
Because they are used so frequently, I suspect they carry enormous amounts of accumulated musical context within the model.

If you listen carefully to a large number of works generated using these kinds of prompts, you may begin to notice recurring similarities.

To me, this suggests that the word “Orchestra” itself has become strongly associated with the dominant musical environments of Suno:
popular music, rock, film scores, game music, and related genres.
In other words, “Orchestra” is not exclusively a classical term.

It is heavily used outside classical music as well.

Film composers use it.
Game composers use it.
Pop ballads use it.

As a result, all of those contexts may become linked together inside the model.

Because of this, I often feel that simply prompting Orchestra tends to generate music that leans toward cinematic or pop-oriented orchestral writing.

This is not necessarily reflected in instrumentation.
The music may indeed use orchestral forces.

However, certain techniques commonly heard in classical repertoire—frequent woodwind ornamentation, pizzicato string writing, brass chorales, and similar gestures—seem less likely to appear unless they are specifically requested.

Harmony is perhaps the clearest example.
When I ask Suno to generate orchestral harmony without further guidance, the results often seem strongly influenced by harmonic conventions derived from pop music and film scoring.

To be clear, this is not a criticism.
I am not suggesting that these results are of lower quality.

Quite the opposite.

Overall quality has improved dramatically.

The question is different:

How can we move beyond those default tendencies when we want something more individual?

This is where orchestral personas become interesting.
The method itself is surprisingly simple.
Instead of using only generic labels, try specifying the formal name of an actual orchestra.

For example:

• Wiener Philharmoniker

• Chicago Symphony Orchestra

• NHK Symphony Orchestra

Simply inserting names like these into a prompt may subtly alter the result.
My impression is that such names introduce musical associations that are not entirely contained within the broader cinematic-orchestral context.

The differences are not large.

That point is extremely important.

In my experience, AI music systems are generally designed to avoid sudden, extreme shifts.

Furthermore, all of these ensembles are still orchestras.

They remain connected to the larger category represented by the word Orchestra itself.

Naturally, some overlap remains.
Yet I do not believe they are identical.
If an orchestra possesses a distinctive sonic identity, traces of that identity may still influence the generated result.

For readers familiar with classical music history, this may be easier to understand.

The Wiener Philharmoniker, for example, is famous for the warmth and softness of its string sound.

When I use that name, I often feel that similar qualities appear more frequently.

The Chicago Symphony Orchestra, known for precision, power, and especially its formidable brass section, seems particularly effective in more dramatic or energetic works.

One of my more recent discoveries has been the NHK Symphony Orchestra.
Whenever I use that name, the resulting orchestral texture often feels remarkably stable and surprisingly compatible with Japanese musical aesthetics.

I have found it particularly effective for orchestral accompaniments in J-Pop-inspired works or music influenced by Studio Ghibli-style soundtracks.

Of course, these are personal observations rather than proven facts.
But they have been consistent enough to keep me experimenting.

The truth is that there are orchestras all over the world.
I have certainly not tested all of them.

Nor do I know which ones were included in Suno’s training data.
However, there are undoubtedly many more worth exploring.

Even among the orchestras I have personally used, there are numerous examples that I simply do not have space to discuss here.
What matters is the principle.

I believe that originality often emerges through the accumulation of many small choices.

Each individual difference may be subtle.
Some listeners may not hear it at all.
Yet these small variations can gradually combine into something uniquely personal.

In that sense, personas are not magic.
They do not transform music overnight.
Rather, they may serve as tools that gently guide our work toward a more individual artistic identity.

Combined with the other techniques discussed in previous articles—such as specifying instrumental behavior, articulation, and detailed performance instructions—they may offer another pathway toward deeper musical expression.

That is all for this installment.
Thank you very much for reading.
In the next article, I would like to move one step further and explore the personas of choirs, early-music ensembles, chamber orchestras, and other specialized performing groups.