VibeVoice LoRA Finetuning
What it is
A repository providing a pipeline to fine-tune the VibeVoice model using Low-Rank Adaptation (LoRA). It supports training with both Hugging Face and local JSONL datasets, handling dual losses for text tokens and acoustic latents.
The problem it solves
Provides a structured way to handle the complexities of training VibeVoice, including dual loss calculations and automatic voice prompt generation.
What you could use it for
- Fine-tuning VibeVoice 1.5B or 7B models for high-quality speech synthesis.
- Training a voice clone using 24kHz audio datasets with optional voice prompts.
- Fine-tuning a model to handle multi-speaker dialogue with specific voice identities.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.