MML vs. GEMS: Yuzo Koshiro's Workflow

Hello! Welcome to the third lesson in our module on the tools and legacy of Genesis-era music.

In the past two lessons, we've taken a deep dive into the two dominant, yet philosophically opposed, music creation workflows for the Sega Genesis. First, we explored the programmer-centric world of Music Macro Language (MML), a text-based system that gave composers direct, low-level control. Then, we examined the GEMS driver, a MIDI-based system designed by Sega of America to provide a more accessible, musician-friendly workflow.

Today, our goal is to bring these two threads together. The learning outcome for this lesson is to compare the creative possibilities and constraints of MML-based workflows versus the GEMS system, using Yuzo Koshiro's MML-driven work as a case study. We will analyze how the choice between these tools wasn't just a technical decision, but a creative one that fundamentally shaped the sound of the Genesis library.

1. The MML Paradigm: Yuzo Koshiro's Custom Toolkit

To understand the zenith of the MML-based approach, we must look at Yuzo Koshiro. His background, combining musical talent with programming skills honed at Falcom on the PC-88 computer, made him uniquely equipped to push the Genesis hardware far beyond its perceived limits. He didn't just use an existing MML; he created his own bespoke environment.

Let's start by reading an excerpt from an interview where Koshiro describes his process.

Koshiro's key insight was that the Genesis's sound subsystem was remarkably similar to the PC-88 he was already an expert on. Both used a Zilog Z80 CPU as a sound controller, and the FM synthesis chips were related. This allowed him to:

1. Compose in a familiar environment: He continued to write music on his PC-88.
2. Port directly: He created a data converter to move his compositions to the Genesis.
3. Customize the driver: He didn't just hand over music data; he handed over source code for his sound driver and specified the commands he needed the programmers to implement.

This is the MML philosophy taken to its logical extreme: the composer is also a systems architect. This deep integration gave him an unparalleled level of control. In the next excerpt, he contrasts this directly with the more standardized MIDI workflow that GEMS is based on.

This is the core of our case study. Koshiro's statements reveal the immense creative possibilities of his MML-based workflow:

• Deep Customization: He could "deeply customize every aspect, from the sounds themselves to the construction of the musical phrases." This means he wasn't limited by the features of a pre-existing driver. If he imagined a new type of arpeggiator or modulation effect, he could program it.
• Hardware Exploitation: His claim of making a 3-voice chip sound like 6 voices points to advanced programming tricks, likely involving rapid channel switching or parameter updates that a standardized system like GEMS would not support.
• Algorithmic Composition: The most striking example is his "self-composing" program for Streets of Rage 3. He applied concepts from object-oriented programming (like C++) and generative systems (like Max/MSP) to create music. This is a level of procedural generation and experimentation that was simply impossible within the GEMS framework. It was a fusion of coding and composition.

The constraint, of course, was the immense technical barrier. This workflow was only possible because Koshiro was an accomplished programmer.

2. The GEMS Paradigm: Abstraction and Accessibility

Now, let's contrast Koshiro's bespoke system with GEMS. As we saw in the last lesson, GEMS was created to solve the exact problem that Koshiro's unique skillset bypassed: the difficulty of programming FM synthesis.

Let's review the design philosophy behind GEMS.

GEMS represents a trade-off. It abstracts away the low-level complexity of the hardware, offering a more intuitive, MIDI-based interface.

The creative possibilities of GEMS were significant:

• Accessibility: It opened up Genesis development to a wider pool of musicians who were comfortable with MIDI sequencers but not with assembly programming.
• Rapid Workflow: The WYSIWYG ("What You Hear Is What You Get") nature of GEMS, where composers could hear their changes in real-time on the actual hardware, dramatically sped up the creative process compared to the MML code-compile-test cycle.
• Powerful Built-in Features: As we learned previously, GEMS had sophisticated, pre-built systems for channel priority management and interactive music via "Mailboxes." This allowed for complex dynamic soundtracks without requiring the composer to program the logic from scratch.

However, this abstraction also introduced constraints:

• Limited by the Driver: A composer using GEMS was ultimately limited to the features implemented in the GEMS driver. They couldn't invent a new form of synthesis or a custom algorithmic process like Koshiro did.
• Potential for Homogeneity: The reliance on a shared set of preset patches (though editable) often led to many GEMS-based soundtracks having a similar sonic character.
• Indirect Hardware Control: The layer of abstraction between the MIDI file and the hardware meant that direct, cycle-accurate hardware tricks were generally not possible. The priority system, for example, is a managed solution to channel limits, not a composer-programmed exploitation of them.

3. Head-to-Head Comparison

Let's summarize the comparison in a table to make the distinctions clear.

Feature / Aspect	Yuzo Koshiro's MML Workflow	GEMS Workflow
Core Philosophy	Composition as programming. Direct, low-level control over hardware.	Composition as music production. Abstraction of hardware complexity.
Primary Interface	Text-based code (custom MML, assembly) in a text editor.	Graphical MIDI sequencer (e.g., Cakewalk) on a PC.
Workflow Cycle	Code -> Compile -> Transfer to dev kit -> Listen. (Slow, iterative)	Compose in MIDI -> Hear in real-time on Genesis. (Fast, WYSIWYG)
Sound Design	Unlimited. Direct register manipulation of the YM2612. Patches are lines of code.	Bounded. Graphical patch editor and preset libraries. Patches are data structures used by the driver.
Expressiveness	Infinite flexibility. Custom-programmed arpeggiators, LFOs, envelopes, and even generative algorithms.	High flexibility. Controlled via standard MIDI messages (CCs, pitch bend) mapped to driver functions.
Hardware Exploits	Encouraged. Enabled novel techniques like procedural music and making the hardware perform beyond spec.	Discouraged/Impossible. The driver manages hardware interaction, preventing direct, unorthodox manipulation.
Target User	Programmer-musician hybrid. Requires deep technical knowledge of the hardware.	Musician/Composer. Requires knowledge of MIDI sequencing, but not programming.

Conclusion

This lesson highlights a fundamental dichotomy in creative tool design. There is no "better" system; there are different systems for different goals and different creators.

Key Takeaways:

• MML-based workflows, exemplified by Yuzo Koshiro's custom system, offered unparalleled creative possibility for those with the requisite programming skills. They allowed for deep hardware exploitation and true innovation, treating the sound chip as a programmable synthesizer. The primary constraint was the extremely high technical barrier to entry and a slower, less interactive workflow.
• The GEMS system prioritized accessibility and efficiency. Its MIDI-based, real-time workflow empowered a generation of Western composers to work on the Genesis. Its built-in features for interactivity and channel management were powerful and streamlined development. The main constraint was the abstraction layer, which, while helpful, limited composers to the features provided by the driver and prevented the kind of radical experimentation seen in Koshiro's work.

The path a composer chose—or had chosen for them—had a profound impact on the music they could create. Koshiro's gritty, dynamic, and technically bewildering scores for Streets of Rage are a direct product of his MML-driven, code-first approach. Conversely, the polished, effective, but often more conventional-sounding scores of many Western titles are a product of the efficient and accessible GEMS pipeline.

In our next lesson, we will broaden our focus and analyze the evolution of FM patch design and timbral complexity from early to late-era Genesis soundtracks. We will see how, regardless of the tools used, composers' understanding of the YM2612 deepened over the console's lifespan, leading to ever more sophisticated and varied sounds.