Voice Conversion

The Voice Change audio modifier changes recorded speech with an installed Pneuma voice model.

Install voice models

For a local server, place .safetensors or legacy .pth models in server/models. You can choose a different directory with SHRIMPLY_PNEUMA_MODEL_DIR.

When using Compose, place models in server/.docker/pneuma/models. Only installed models appear in Shrimply; training a voice model is not part of the server.

Change a voice

1. Select an audio clip containing speech.

2. Add the Voice Change audio modifier.

3. Choose one of the models reported by the compute server.

4. Preview the clip and adjust pitch, speed, or the F0 method if needed.

Keep Maintain pitch while changing speed enabled when changing speed should not also shift the voice’s pitch.