Improve Procedural Audio Mix and Role Mapping (melody clarity + depth)

[joenb33]

Improve procedural mix + role mapping clarity

Hi! First of all, thanks for this great project. It is very fun to use and it produces a lot of good sounding music.

I noticed that in procedural mode, some MIDI files (especially simple/game-style files) can sound:

• too low/muddy
• missing clear lead notes
• flatter than expected

I tested a few focused changes locally that improved this a lot.

What seems to cause it

1. Role overwrite in synthesis setup
   When multiple assignments share the same role, later assignments can overwrite earlier ones.

2. No melody fallback in some mappings
   Some files end up mostly as texture (or with weak lead presence), so the main tune is not clear.

3. Mix balance favors low/soft layers
   Bass/texture can dominate while melody is too soft/dull.

Suggested fixes

1. Merge same-role assignments in src/synthesis/SynthesisEngine.ts before layer setup
   (combine events/chords instead of last-write-wins).

2. Add melody safeguard in src/core/RoleMapper.ts
   If no melody is assigned but tracks exist, promote the best melody candidate above a threshold.

3. Rebalance synth mix in src/synthesis/SynthesisEngine.ts
   Slightly reduce bass dominance and make melody more present/bright.

Result (local testing)

• Better melody clarity
• Less “missing notes” feeling
• More consistent depth across mixed MIDI quality

[b1rdmania]

Thanks a lot for this — genuinely great breakdown.

Integrated suggestions 1 and 2: reworked the role mapper (RoleMapper.ts) so same-role tracks get merged instead of overwriting each other, and added a melody safeguard so the tune keeps a clear lead when there's one to find. Shipped in v1.7.1. Seems to have helped.

On number 3 — I completely agree, I've noticed the same muddiness on a few tracks recently. For now I'm going to park it though. It's a global mix change and I don't want to shift the sound people already expect from v1 without more signal. Might have another look next month.

In the meantime: I actually built a v2 sound engine. I noticed, with some horror, about a month after launching this, that v1 sounds way more like a Super Nintendo than a Game Boy. So I chipped away at a different version that goes for the actual Game Boy chip (real LFSR noise, duty cycles, wavetables). It's live and still rough — have a look at https://www.wario.style/v2.html if you fancy it and see if you can pick any errors. #5 has the open problem (mapping a full MIDI onto 4 fixed voices) if you're up for a crack.

Thanks again — really good stuff.

[b1rdmania]

PS — shipped this as v1.7.1 and gave you a shoutout in the version history on the site. Cheers again.