From 60f88737184f67c0cc116fc62651cd2566d2f752 Mon Sep 17 00:00:00 2001 From: mrfakename Date: Tue, 23 Sep 2025 08:50:07 -0700 Subject: [PATCH] Update FINETUNING.md --- FINETUNING.md | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/FINETUNING.md b/FINETUNING.md index eee9832..ac4bb65 100644 --- a/FINETUNING.md +++ b/FINETUNING.md @@ -1,12 +1,17 @@ # Finetuning -> [!IMPORTANT] -> Finetuning is still **very experimental** and not well tested yet! - More instructions coming soon. +VibeVoice finetuning works wonders - both for adaptipng VibeVoice to new languages and for better voice cloning of a single voice. + Join the [Discord](https://discord.gg/ZDEYTTRxWG) for support. Also take a look at [voicepowered-ai/VibeVoice-finetuning](https://github.com/voicepowered-ai/VibeVoice-finetuning). +## Notes + +* Members of the community have observed that for fine-tuning VibeVoice on a single voice, it is often beneficial to set `voice_prompt_drop_rate` to `1.0` and avoid use of voice cloning/reference audio all together during inference. This can lead to more natural speech generation. **If you are training on a single speaker I highly recommend you try this.** + +## Example Script + Example script: ```bash