Java voice add-on

Verity JE Extra: Custom Voice, Fish Audio, and Setup

Verity JE Extra is a third-party add-on for Verity JE, not a replacement for the mod. Use it to change voice or chat providers after your base Verity installation works.

Information checked 2026-10-01 Dates and versions can change.

What Verity JE Extra changes

Verity JE Extra adds optional voice and chat settings to Verity JE on Minecraft 1.20.1 with Forge. It is a separate project by somewhatmaxy. Start at the author’s Modrinth listing, not a repack advertised as a complete Verity installer.

The useful distinction is between three jobs: writing an AI response, reading that response aloud, and recognizing what you say into a microphone. Changing one does not necessarily change the others. Decide which job you want to improve before installing another component.

The project documents custom chat endpoints, cloud speech including Fish Audio, and offline Piper voice models. Its overrides start disabled. A separate Host Mode is experimental. These are optional additions; a working base installation does not need them simply because they are newer.

If you only want to start a normal conversation, use the Java installation guide first. This page covers choosing an add-on configuration, checking what actually changed, and undoing an unsuccessful test without losing a working game profile.

Check compatibility before changing your installation

The add-on requires Verity JE 6.0 or newer and YACL. That excludes the base project’s 5.7.4 stable release. The base project’s current CurseForge listing also offers 6.0.0-beta.9 for Forge 1.20.1. A file being available does not mean you should replace a stable profile that already does what you need.

Use this decision before downloading:

Your current setupNext step
Verity 5.7.4 works and you do not need custom voiceKeep your existing setup
Verity 5.7.4 works but you want JE ExtraCreate a separate 6.x test profile
Verity 6.x works with Forge 1.20.1Compare the add-on release requirements with your installed dependencies
NeoForge 1.21.1, Fabric, or BedrockDo not assume this add-on is compatible
Verity fails before reaching a worldFix the base installation before introducing overrides

The beta.4 release notes explain support across the differently named Verity 6.x releases. They also identify a configuration-screen crash with older YACL builds and recommend 3.6.6 or newer. That is a specific troubleshooting note, not an instruction to install an arbitrary YACL file for another loader or Minecraft version.

Write down four values from your launcher: Minecraft, Forge, Verity JE, and YACL. Add JE Extra’s exact release when you install it. “Latest” is not a useful record when two people downloaded their files on different days. Keep this small version list with your test profile so a later update can be compared precisely.

Install without disturbing a working world

Use a separate launcher profile or a copy of your existing instance. Back up important worlds before opening them with a different Verity branch. Keep the backup outside the test instance so removing the instance cannot remove the only copy.

  1. Start the compatible base Verity profile without JE Extra. Enter a disposable world and try a short typed conversation. Record whether a text reply and audible reply both work.
  2. Close Minecraft completely. Install the compatible add-on through its project listing into that same test profile. Retain the base mod and its required dependencies; JE Extra does not replace them.
  3. Launch again and open Verity’s configuration screen. Look for the Verity JE Extra tab. Before enabling an override, repeat the same short conversation.

That second conversation gives you a useful comparison. If the game now crashes before any override is enabled, changing a voice model will not address the startup failure. Check the dependency and release combination. If the game loads but the tab is missing, confirm that you launched the instance you modified, not another profile with the same display name.

Do not install several copies of the add-on while trying to make the tab appear. Compare the actual mod list against the files you intended to use, then remove only the obsolete duplicate from the test profile. Keep the original working installation untouched until the new one passes your checks.

Choose the smallest change that solves your problem

For a different spoken voice, leave the conversation provider alone initially. For a different AI model, leave speech alone. A useful first test has one changed setting and one clear result: the same prompt now returns text, or the same answer is spoken in the intended voice.

Keep a short private note with the setting changed, the previous value, and the outcome. Do not put credentials into that note. If a test fails, turn off the changed override before trying another. Changing both chat and cloud speech together makes a silent answer ambiguous: either no response was generated or a generated response could not be spoken.

Use a harmless short prompt such as asking for one building material. Avoid personal information while testing a new service. Keep test sessions brief when using a paid provider, and check its usage dashboard before inviting friends to repeat requests.

Use a local Piper voice

The author’s Offline Voice instructions use Import .onnx… with the matching tokens.txt and espeak-ng-data alongside the model. Keep the supplied voice assets together. This is not a promise that any file with the same extension will work; use a model format supported by the add-on’s current instructions. The project page also links a TTS setup video for following the interface visually.

Before importing, inspect what you actually downloaded. An archive still needs to be extracted, and nested folders can leave the selected model separated from its supporting files. Preserve the bundle’s filenames. Randomly renaming another file to match a missing filename does not supply the missing model data.

After importing, test a short response you can easily recognize. Note whether text appears, whether audio begins, and whether the voice sounds like the chosen model. A visible answer with no sound points you toward speech or playback checks; no text answer is a different problem. Return to the previously working speech configuration if the import cannot be loaded.

Offline speech is not the same as offline conversation. To avoid online processing throughout a conversation, you would need to check the chat model and any speech-recognition service separately. The Ollama guide covers local chat configuration; the speech guide separates microphone input from voice output.

Keep voice licensing separate from technical compatibility. A voice may load successfully without giving you permission to redistribute its files or imitate a real person publicly. Use voices you are entitled to use, and check their terms before including them in a shared modpack or public recording.

Configure Fish Audio without replacing chat settings

The Fish Audio option uses its own key and voice model ID. The add-on supplies the service endpoint and engine selection for that layout. Follow the current add-on instructions rather than copying a generic cloud-voice configuration intended for another service.

Fish Audio’s text-to-speech API documentation distinguishes authentication, the speech engine, and the reference voice. A reference voice ID is not your account credential. Keep those values in their corresponding fields, and obtain them from your own provider account.

For the first check, keep a working text-chat setup unchanged and enable only the intended speech override. Request a short answer. If the text arrives but audio does not, inspect the speech error and the Fish Audio account before regenerating your chat key. A successful chat request does not prove a separate speech account is authorized or funded.

Do not repeatedly submit long prompts to test the same failure. Read the response code, check account access and usage, then retry once after a specific correction. If the provider works separately but the add-on still fails, include both results in a redacted bug report. That comparison is more useful than reporting only that Verity is silent.

Understand which settings still belong to Verity

The add-on does not replace speech-to-text. Microphone selection, permission, and recognition problems therefore still need the base mod’s input settings. Also, the author notes that interruption does not immediately stop ongoing audio. Do not treat that known limitation as proof that Minecraft has frozen.

For custom chat, the project offers an OpenAI-compatible endpoint override. Compatibility is about the request interface, not just putting a provider’s homepage into a URL field. Use the provider’s documented API address and model identifier. Do not paste an account login link, billing page, or website search URL into the endpoint field.

When debugging, note whether the add-on override is enabled. An edited value in the base mod may have no effect on the request you are currently testing if an override controls it. Turn the relevant override off, repeat the same prompt, and compare. That gives you a reversible test before deleting configuration or reinstalling the game.

Host Mode: test with one friend first

The author describes Host Mode as a way for the host to generate speech once and share the audio with nearby players. It is explicitly still being tested. Treat a private two-player session as a compatibility check, not a guarantee for a public server or paid hosting plan.

Before enabling it, establish that both players can join the world and use the base mod. Then compare one short response with the feature off and on. Ask each player separately whether they saw the text, heard sound, heard the same voice, and experienced an unusual delay. Keep players nearby for the initial check so distance does not add another unknown.

The project’s claim that credentials stay on the host is not a reason to share configuration files. Do not assume that a dedicated server with no interactive audio device behaves like a hosted single-player world. The multiplayer guide explains the host questions to settle before extending a local setup.

Report a problem or return to the base mod

The author’s GitHub issue tracker names config/verityjeextra.json as a file containing credentials. It also warns that endpoint URLs can identify users. Do not publish a complete configuration or unreviewed log. Redact credentials, private endpoints, personal paths, and conversation content before attaching anything.

Report the five version values, which override was enabled, the action that failed, and whether it works with that override disabled. Include a short error excerpt, not an entire private session. If removing JE Extra from the test profile leaves the same problem, investigate the base mod or provider rather than assigning the failure to the add-on automatically.

To return to your earlier setup, close the game and use the backed-up working profile. Do not open your only important world through repeated upgrade and downgrade attempts. Once a test configuration works, keep its version list and add further changes one at a time. A reproducible working combination is more useful than a folder containing every release you tried.

Quick answers

Frequently asked questions

Does Verity JE Extra work with Verity 5.7.4?

The add-on lists Verity JE 6.0 or newer as a requirement. Do not install it into a working 5.7.4 profile. Keep that profile and test a compatible 6.x installation separately.

Does Offline Voice make all of Verity work offline?

No. Offline Voice changes speech output. Chat generation and microphone transcription use separate settings and may still contact online services.

Why does opening the configuration screen crash Minecraft?

The beta.4 release notes identify older YACL versions as one cause and recommend YACL 3.6.6 or newer. Choose a compatible Forge 1.20.1 build and check the crash report rather than assuming every configuration crash has the same cause.

Can JE Extra fix Verity not hearing my microphone?

Its documented overrides do not replace speech-to-text. If typed conversation works but microphone input does not, check Verity's own speech input settings and device permissions.

Is Host Mode guaranteed to work on a dedicated server?

No. The author describes Host Mode as new and still being tested in multiplayer. Confirm support for your exact build and hosting arrangement before relying on shared voice generation.

References

Where this information comes from

Project page

Verity JE Extra by somewhatmaxy

Third-party Forge 1.20.1 add-on for Verity JE 6.x; optional chat and TTS overrides, Piper and Fish Audio setup, unchanged speech input, and experimental Host Mode.

Checked

Project page

Verity JE Extra beta.4 compatibility notes

Configuration-screen crash with older YACL versions; the release notes recommend YACL 3.6.6 or newer and describe compatibility changes across Verity 6.x builds.

Checked

Project page

Verity JE on CurseForge

Java project details, Project ID 1591438, stable Forge 1.20.1 main file 5.7.4, 6.0.0-beta.9 listing, legacy NeoForge branch, AI options, gallery, and author notices.

Checked