PopotoVox

"Every popoto gets a voice."

A self-contained FFXIV Dalamud plugin that gives every NPC a distinct, locally-cast synthetic voice. No cloud, no paid services, no dependency on other plugins.

By kernelcurry.

Status

Released and installable — v0.9.x ships via the custom Dalamud repo (install steps below), and a submission to the official Dalamud plugin repository is in review. Full pipeline: dialogue capture → NPC identity resolution → rules + local-LLM casting → local TTS, with a management UI (NPC browser, bio cards, override editor, presets) and a signed, SHA-256-verified asset downloader.

Three selectable voice engines (a quality ladder, chosen via Settings or the Low/Medium/High/Ultra presets):

Engine	Preset	Hardware	Notes
Piper	Low / Medium	CPU	Lightweight/fast, 904 voices, more robotic (Medium adds smart casting)
Kokoro (default)	High	CPU	Natural, 53 voices across 9 real-world accents, no GPU needed
VoxCPM2	Ultra	NVIDIA GPU	Most lifelike — one model that designs a unique voice per NPC (in the speaker's native tongue, for a real accent) then clones it to perform each line with directed emotion; streaming playback so speech starts almost immediately; cached per NPC. The heaviest tier: a large one-time download (GPU runtime + model)

A small local LLM ("smart casting") picks a fitting voice per NPC and, on Ultra, directs restrained mood-appropriate emotion per line. First run launches a guided setup wizard (choose a voice → optional smart casting → download → done); or open /popotovox → Storage and install an engine, then talk to NPCs.

Ambient world chatter (optional, off by default): beyond interactive dialogue, PopotoVox can voice the overhead speech bubbles NPCs say spontaneously as you walk by. These fade with distance using a realistic acoustic falloff (modelled on how real speech attenuates) and can swell and fade in real time as you move past the speaker — so a busy hub feels alive. Multiple bubbles can play at once through a real audio mixer, each at its own distance volume; during a conversation, nearby chatter politely ducks rather than going silent. Tunable hearing distance and overlap limit, and matched to each NPC's voice like every other line.

The management UI has a modern, card-based NPC browser that shows each character's identity, full gear, and a transparent "why this voice" breakdown (in the casting director's own words). It shows one card per character, even though the game tracks the same NPC under many internal IDs (a new one per quest scene — story characters can span dozens); the card notes how many IDs it folds together, and overrides/re-casts apply to all of them at once. Nameless passers-by are listed with readable descriptions ("Unnamed Hyur woman — The Doman Enclave") behind a "Show unnamed" filter.

Repo layout

The release repository for the plugin — the minimum needed to build, verify and install it. (Development happens elsewhere; this repo tracks what ships.)

• plugin/ — the Dalamud plugin (C#, .NET 10, Windows; built via Dalamud.NET.Sdk/15.0.0).
• tts-host/ — the isolated TTS helper process (Kokoro via sherpa-onnx) the plugin launches.
• voxcpm-host/ — the isolated VoxCPM2 (Ultra) Python host script.
• docs/LICENSES.md — third-party dependency ledger.
• images/ — the plugin icon.
• Dockerfile, build.sh — Docker-based build (no host .NET SDK needed).
• .github/workflows/build.yml — CI: builds on every push, releases on v* tags.
• repo.json — Dalamud custom-repo manifest.

Install in-game (preferred — via Dalamud custom repo)

This is the one-click path. You add a URL to Dalamud once and the Plugin Installer treats PopotoVox like any other plugin.

1. In FFXIV: /xlsettings → Experimental → Custom Plugin Repositories.
2. Add this URL and check the enabled box:

   https://raw.githubusercontent.com/kernelcurry/popotovox-plugin/main/repo.json
   

3. Save and close. Open the Plugin Installer, find PopotoVox, install.

Custom-repo installs only work once a GitHub release exists. The CI workflow publishes a release whenever a v* tag is pushed (e.g. git tag v0.9.0 && git push origin v0.9.0).

Build locally (no .NET SDK required, uses Docker)

If you'd rather build from source — for example to test an unreleased commit — use the Docker build. You only need Docker installed.

./build.sh

Output: ./dist/PopotoVox/ containing PopotoVox.dll, PopotoVox.json, and dependencies.

Then in FFXIV:

1. /xlsettings → Experimental → Dev Plugin Locations.
2. Add the absolute path printed by build.sh (e.g. /path/to/PopotoVox/dist).
3. Plugin Installer → Dev Tools tab → enable PopotoVox.

The Docker build downloads Dalamud reference DLLs from the official distribution channel and targets net10.0-windows from a Linux container via EnableWindowsTargeting. No Windows host required.

Build locally (with .NET 10 SDK)

If you have the SDK installed and prefer a native build:

cd plugin
dotnet restore -p:EnableWindowsTargeting=true
dotnet build  -p:EnableWindowsTargeting=true

EnableWindowsTargeting=true is only needed when building on non-Windows hosts (Linux / macOS) — on Windows it's a no-op. Dalamud reference DLLs are resolved by Dalamud.NET.Sdk from:

• Windows: %AppData%\XIVLauncher\addon\Hooks\dev\ (default)
• Linux: $HOME/.xlcore/dalamud/Hooks/dev/ (default)
• Override anywhere: set DALAMUD_HOME=/path/to/extracted/dalamud-distrib/

In-game commands

Command	Action
/popotovox (alias /pvox)	Open the main window (NPCs)
/popotovox config	Jump to Settings
/popotovox setup	Reopen the first-run setup wizard
/popotovox diag	Jump to the System dashboard — live pipeline activity, status, and debug info

License

GPL-3.0 (see LICENSE). Free and open-source for everyone, forever. The default voice engine (Kokoro via sherpa-onnx) links GPL-3.0 espeak-ng, so the combined work is GPL-3.0. See docs/LICENSES.md for the full dependency audit — we still avoid all non-commercial / research-only licenses.

Author

kernelcurry · info@kernelcurry.com