Bilingual EVA Announcer: Letting Custom Voice Take Over from Vanilla

Date: 2026-08-07

1. Background

The engine’s EVA announcer voice is part of the game experience. Vanilla plays ceva/csof wav from mix via the [DialogList] table in eva.ini. Making the announcer speak Chinese or custom English requires two things: first, knowing exactly which EVA events the engine can trigger; second, designing a mapping structure for custom voice.

2. Mapping Out the Engine’s 64 EVA Events

Don’t guess — dig from source and vanilla data:

Engine source eva.play("EVA_xxx") call sites: grep all literals, dedupe to get the EVA events the engine actually triggers;
Extract eva.ini from ra2.mix > local.mix, parse [DialogList] (356 valid entries), and read Text (vanilla English) and Allied/Russian (voice files) per entry;
Cross-check: all 64 engine-referenced events exist in eva.ini with Text + voice.

Key insight: the engine only triggers 64 events; the mission-objective classes in the 356-entry full list (EVA_MissionAccomplished etc.) are unused. So custom voice only needs the subset the engine actually fires.

3. Event Semantics Annotation

The 64 events were grouped by purpose to decide what to make:

Group	Count	Notes
Economy/build/unit (core)	30	construction complete, insufficient funds, low power, unit ready
Capture/garrison	4	building captured, garrisoned, abandoned
Spy/infiltrate/upgrade	12	cash stolen, radar sabotaged, unit promoted
Alliance	5	formed, broken, requested
Superweapon	13	nuke, Iron Curtain, chrono, lightning (optional, MOD has none)

Each entry has vanilla English + Chinese translation, forming a bilingual table.

4. Voice Mapping Structure (JSON + Schema)

design/voice/{en,zh-CN}/event_voice_mapping.json:

{
  "events": {
    "EVA_ConstructionComplete": {
      "filename": "EVA_ConstructionComplete.wav",
      "text": "建造完成。"
    }
  }
}

schema/voice.schema.json locks down:

Event keys must belong to the 64-event EVA enum (propertyNames.enum);
Each entry requires filename + text, additionalProperties:false;
filename must equal the event key + .wav.

Typos fail at edit time instead of runtime.

5. Engine Consumption Side (Pending)

Current state: voice files are generated and injected into OPFS; the mapping is in place and passes Schema validation. But the engine’s eva.play() still goes through eva.ini → vanilla mix voice. Making custom voice actually play requires engine wiring: event trigger → read design/voice/{lang}/mapping → play custom wav (fall back to eva.ini vanilla if missing). This is the next candidate task, recorded in docs/PORTING.md.

6. Conclusion

The bilingual announcer skeleton is “event enum (64) + bilingual text + JSON Schema mapping + TTS generation”. Map the engine’s trigger surface first, so you don’t synthesize 300 unused lines out of the 356-entry full set; the Schema locking the enum keeps the mapping in sync with engine reality — the first step toward letting vanilla voice step aside.

This article is licensed under Mulan Open Works License Attribution-PatentLicensed, Version 1 (MulanOWL BY-PL v1), or alternatively CC0-1.0 (public domain dedication). Copyright reserved. Attribution required. For enterprise compliance, please retain the original license notice. Original version: https://k4m7v2pz.github.io/en/blog/games/openyrweb-bilingual-eva-voice.html

Keyboard shortcuts