voxcpm-voice

voxcpm-voice

Ship the next thing you're building with a full cast of voices you own — no voice actors, no studio time, no recording sessions. By the end of today you'll have characters, narrators, announcers, and AI personalities saved in a local library that you can call on any time, for any line, and hear back the same voice every time. Your game, app, video, or podcast gets a voice cast that's ready whenever you are.

You'll describe what you want in plain English and get a working voice in under a minute. You'll pick the takes you love and keep them forever. You'll hand over a script of 500 lines and watch them render in the same character you saved last week. If you have a recording of a real voice — your own, a friend's, a voice actor's demo reel — you'll turn it into a reusable asset in seconds.

What you'll walk away with

A voice cast that's uniquely yours. Not stock TTS that sounds like every other app. Characters you designed or cloned, each one distinct, each one a permanent asset you control.

The same voice, every single time. Once a voice is in your library, every line you generate in it sounds like the same speaker. No drift across a session. No drift across months. Voice number 17 in your script sounds like voice number 1.

Speed you can build a project around. A voice in a minute. A voiceline in seconds. A 200-line VO bank in a coffee break. No retakes, no scheduling, no emails chasing files.

Your own voice (or anyone's) as an asset. Drop in a clean 15-second recording and the voice is yours to use forever. Add a transcript for the highest fidelity reproduction.

Delivery control when you need it. "Urgent, hushed." "Slower, menacing." "Cheerful, fast." You tell the line how to be delivered without changing who's speaking.

A library that compounds. Every voice you add, every line you ship, sticks around. Your next project starts with everything you built in the last one.

Outputs that live inside your project. Run one init command and your rendered VO lands in a vo/ folder next to your code — check the scripts into git, ship the audio with your builds, keep it all version-controlled.

Get set up

claude plugin marketplace add Sortium-io/voxcpm-voice
claude plugin install voxcpm-voice

Restart your Claude Code session. Your first voice will auto-install the backing model (~2 GB, one-time). From then on, you're in seconds-per-voice territory.

Works on macOS, Linux, and Windows. If nvidia-smi is on your PATH (Linux or Windows with an NVIDIA card), the installer pulls CUDA 12.1 PyTorch wheels automatically — generation is dramatically faster on GPU. Apple Silicon uses MPS; anything else falls back to CPU (still works, just slower).

You'll need: Claude Code, Python 3.10+, ~3 GB free disk. GPU helps but isn't required.

Updating or starting over

claude plugin uninstall voxcpm-voice && claude plugin install voxcpm-voice   # update
claude plugin marketplace remove voxcpm-voice                                  # forget the marketplace
rm -rf ~/voxcpm-voice                                                          # nuke the library + venv

Your first voice, in about a minute

After installing, restart Claude Code. The skill registers on the next session. Then open a Claude Code session and say:

Use voxcpm-voice. I want a gruff male drill sergeant in his fifties —
weathered gravelly baritone, shouted parade-ground commands, hoarse edge.
Give me 3 takes so I can pick.

Three takes land in ~/voxcpm-voice/voices/Drill_Sergeant/samples/. Play them. When one's the one, say:

Save take 2 as the drill sergeant's reference.

You just built a reusable character. The drill sergeant is now yours, forever, for any future line.

Or: run the entire fresh-install smoke test in one session

Skip the restart by invoking the scripts directly. Paste this whole block into any Claude Code session on a fresh machine — it handles install, venv setup, project scaffolding, voice design, and rendered voicelines end to end, and tells you where the WAVs landed.

Install and smoke-test the voxcpm-voice plugin from the Sortium-io
marketplace on this machine — end to end, no workarounds.

Step 1 — install the plugin:
  claude plugin marketplace add Sortium-io/voxcpm-voice
  claude plugin install voxcpm-voice

Step 2 — find the plugin path. It lands under one of these; discover
the actual one with `ls` / `dir`:
  macOS/Linux:  ~/.claude/plugins/marketplaces/voxcpm-voice/
                ~/.claude/plugins/cache/voxcpm-voice/voxcpm-voice/<version>/
  Windows:      %USERPROFILE%\.claude\plugins\marketplaces\voxcpm-voice\
                %USERPROFILE%\.claude\plugins\cache\voxcpm-voice\voxcpm-voice\<version>\
Use whichever has `skills/voxcpm-voice/scripts/` inside. Call that PLUGIN_ROOT.