Layer one: parent mod and loader
Start with a clean Java profile. For the current Better Voice file, the checked route is Minecraft
1.20.1 Forge. The current file changelog says Better Voice 4.0.0 adds support for Verity JE
6.0.0 and is incompatible with earlier Verity JE versions. The broader project table may still
show older requirement wording, so treat the exact file page as the current compatibility source.
Add only the needed parent mod and dependencies first. Launch the profile and confirm
the game opens before adding extra horror packs, server-side utilities, visual mods, or provider
experiments. This protects you from diagnosing a sound issue when the real issue is a loader,
dependency, or version mismatch.
After the parent Verity JE route loads, test one short typed prompt. If the text response fails,
check Groq, Ollama, OpenRouter, status 401, status 429, local server state, or model behavior.
If the text response works but voice does not, Better Voice, Minecraft audio, TTS provider,
voice ID, internet access, or provider credits become the likely layer.
Layer two: Fish Audio
Fish Audio is the quality and emotion path. It uses its own API key and a voice or reference ID.
Voice mimicking also belongs to this side. If the mimic fails, the checked project notes point to
common causes such as invalid Fish key, insufficient credits, or not enough clean push-to-talk
speech. Do not post the Fish key, the full voice ID, or a private audio sample when asking for help.
Layer three: Cartesia
Cartesia is the speed and pronunciation path. It has its own API key and Voice ID in a separate
config tab. If a player changes providers and Verity becomes silent, confirm that the provider
was actually switched, the right tab has a key and voice ID, Minecraft is online, and the account
has access to the selected voice. Do not reuse a Groq key in a Cartesia field.
Layer four: audio output
When chat works and provider settings look right, check ordinary output. Raise Minecraft master
volume, voice or sound sliders, operating-system output, headset routing, resource packs, and
server/client parity. If the player is on a server, confirm the add-on and dependencies are present
where the feature expects them. A working text response means the problem may be after the AI layer.