TTS

A BepInEx mod that voices every dialogue line in Vanguard Galaxy using neural text-to-speech.

• 2831 pre-rendered OGG clips cover every story and side-quest dialogue line the game currently ships with
• 42+ NPCs get distinct voices from Kokoro v1.0's multilingual speaker bank — Japanese, Hindi, Spanish, French, Portuguese accents where the character calls for it
• ECHO (your ship AI) and the player captain get hand-picked voices, with 6 captain presets (3 male + 3 female) selectable in config
• Live fallback via the bundled Kokoro engine catches anything the pack misses (dynamic commander-name lines, post-patch additions, procedural station names)

Install

1. Install BepInEx 5.x into the game folder — grab the BepInEx_win_x64_5.4.x.zip from the BepInEx releases and unzip it into your Vanguard Galaxy install folder (next to VanguardGalaxy.exe).
2. Launch the game once so BepInEx creates its BepInEx/plugins/ and BepInEx/config/ subfolders, then close it.
3. Download the VGTTS release from Nexus Mods.
4. Unzip into BepInEx/plugins/. The zip contains a single VGTTS/ folder that drops in cleanly:

   VanguardGalaxy/BepInEx/plugins/
     VGTTS/
       VGTTS.dll
       prerender/...     (2831 OGG files + manifest)
       tools/kokoro/...  (neural TTS model)
       tools/sherpa/...  (inference binary)
   

5. Launch the game. Dialogue lines should start speaking.

Config

BepInEx writes the config to BepInEx/config/vgtts.cfg after the first launch. Key knobs:

[General]
Enabled = true                  # master on/off
DialogueTTS = true              # speak NPC dialogue
EchoTTS = true                  # speak ECHO's ambient HUD hints

[Voice]
CaptainPreset = auto            # auto | m1 | m2 | m3 | f1 | f2 | f3

Captain voice presets:

Preset	Voice	Character
m1	Kokoro am_fenrir	rugged American explorer
m2	Kokoro am_onyx	deep, confident
m3	Kokoro bm_fable	British rogue
f1	Kokoro af_alloy	calm, measured
f2	Kokoro bf_alice	British female
f3	Kokoro af_heart	warm flagship
auto	picks m1 or f1 based on your commander's gender

The [Voices] section lists every known NPC with its default Kokoro speaker ID. Override any NPC by changing the value — e.g. Arle = kokoro:3 would swap Arle to the warm flagship female voice. See the full Kokoro speaker list at hexgrad/Kokoro-82M.

Troubleshooting

No audio at all

• Check the BepInEx console (enable via BepInEx/config/BepInEx.cfg → [Logging.Console] → Enabled = true). You should see Kokoro TTS loaded. NPC profiles seeded: 49, prerender entries: 2831, ... on startup.
• If instead you see Kokoro bundle missing, TTS disabled, the VGTTS/tools/kokoro/ or VGTTS/tools/sherpa/ folder didn't unpack correctly. Re-download and re-extract.

Wrong voice / wrong gender for a specific NPC

• Edit the [Voices] section in vgtts.cfg and set the speaker to a different Kokoro SID (e.g. Arle = kokoro:3). Delete the entry entirely and relaunch to restore the default.

Some dialogue lines play with a noticeable delay on first utterance

• That's the live-TTS fallback synthesizing a line the pack doesn't cover. Once synthesized it's cached — the same line plays instantly the second time. Common causes: game patches added new text, commander-name lines during the first ~20 seconds of a save load before the warm-up finishes, station names in procedurally-generated systems.

"Captain Fank" etc. lines don't match your callsign

• The plugin reads your commander's callsign at save load and pre-warms those lines. If you rename mid-save, re-load the save to trigger a fresh warm-up.

Known limitations

• Procedural station names (e.g. "Welcome to The Oracle Spire") are built from random word lists at runtime and can't be pre-rendered. These always use live TTS.
• Conquest rank titles vary with your progression and also stay on live TTS.
• Scottish accent — Kokoro has no Scottish speaker, so Elias McIntire uses a plain American male voice. Tracked as an open gap.
• Game version drift — when the game patches add new dialogue, those lines will miss the pack and fall through to live TTS. Sounds fine but uses the English-Kokoro voice for accent NPCs. To contribute a delta for a new patch, see "Contributing" below.

Releasing (for maintainers)

Creating a GitHub Release auto-builds and uploads the zip via .github/workflows/release.yml. CI compiles VGTTS.dll against a publicized stub of Assembly-CSharp.dll committed at VGTTS/lib/Assembly-CSharp.dll (method signatures only — no game code). The real game assembly takes over at runtime via Mono's late binding.

# One-time per game update — regenerate the publicized stub from your
# current install and commit it. BepInEx-standard tool, MIT-licensed.
# The --strip flag is mandatory: it replaces every method body with
# `throw null;` so the committed file contains only public signatures,
# not the game's IL. Running without --strip reintroduces the copyright
# leak that was purged in the 2026-04-23 history rewrite.
dotnet tool install -g BepInEx.AssemblyPublicizer.Cli
assembly-publicizer --strip \
  "$GAME_DIR/VanguardGalaxy_Data/Managed/Assembly-CSharp.dll" \
  -o VGTTS/lib/Assembly-CSharp.dll
git add VGTTS/lib/Assembly-CSharp.dll && git commit -m "chore: refresh publicized stub for game vX.Y"

# Tag + release
gh release create v1.2.0 --title "v1.2.0 …" --notes-file CHANGELOG-v1.2.0.md

GitHub Actions then: checks out the tag, verifies the stub is present, runs dotnet build, fetches the Kokoro + sherpa-onnx bundles, stages the shipping layout, and uploads VGTTS-v1.2.0.zip back to the release.

To re-run on a failed release without recreating it: gh workflow run release.yml -f tag=v1.2.0.

Contributing

The dialogue pack is built from the game's decompiled source, so when a patch adds new lines, someone has to re-extract and re-render. Workflow:

# 1. Harvest your session's misses
cp "$GAME/BepInEx/cache/VGTTS/unprerendered.tsv" .

# 2. Render them (needs ~400 MB Kokoro bundle)
make download-kokoro
python tools/prerender/render_missing.py --input unprerendered.tsv

# 3. Commit + PR
git add prerender/ && git commit -m "chore: delta for patch X.Y"

Full rebuild from scratch is also supported — see tools/prerender/extract_dialogue.py + tools/prerender/render_missing.py.

Credits

• Vanguard Galaxy by Bat Roost Games — the game all this dialogue comes from
• Kokoro v1.0 (hexgrad/Kokoro-82M) — the TTS model voicing every line
• sherpa-onnx (k2-fsa/sherpa-onnx) — runtime inference
• BepInEx 5 — mod loader
• HarmonyX — runtime patching

See THIRD_PARTY_NOTICES.md for full attribution.

License

MIT. See LICENSE.