---
language: en
license: mit
tags:
  - voice-cloning
  - text-to-speech
  - lora
  - bedtime-stories
  - voxcpm2
base_model: openbmb/VoxCPM2
---

# VoxCPM2 Bedtime Story LoRA

LoRA adapter fine-tuned on VoxCPM2 for bedtime story narration in English.

## Training details
- **Base model**: `openbmb/VoxCPM2`
- **Dataset**: LJSpeech (6,550 clips, ~12h — half-data experiment)
- **Method**: LoRA (r=32, alpha=64, targeting DiT attention layers)
- **Best experiment**: exp6 (r=32, lr=5e-5, half data)
- **Validation loss**: 0.872 (best across 6 experiments)
- **GPU**: A100-80GB via Modal

## How to use

```python
from voxcpm import VoxCPM

model = VoxCPM.from_pretrained("openbmb/VoxCPM2", device="cuda", load_denoiser=True, optimize=True)
# Apply LoRA weights manually or load via adapter
```

## Evaluation results

| Experiment | Config | Train Loss | Val Loss |
|-----------|--------|-----------|----------|
| exp1 | r32, lr=1e-4 | 0.823 | 0.918 |
| exp2 | r16, lr=5e-5 | 0.765 | 0.899 |
| exp3 | r64, lr=1e-4 | 0.838 | 0.908 |
| exp4 | r32, lr=5e-5, 200 steps | 0.803 | 0.884 |
| exp5 | r32, lr=2e-4 | 0.832 | 0.896 |
| **exp6** | **r32, lr=5e-5, half data** | **0.935** | **0.872** |

## Part of DreamVoice
This model is part of [DreamVoice](https://huggingface.co/spaces/build-small-hackathon/dreamvoice) — bedtime stories in a parent's cloned voice.