README.md

---
license: other
tags:
  - audio
  - tts
  - vae
  - vocoder
  - safetensors
  - dramabox
  - resembleai
base_model: ResembleAI/Dramabox
---

# Dramabox — Audio VAE + Vocoder

This repository contains a **merged safetensors checkpoint** extracted from
[ResembleAI/Dramabox](https://huggingface.co/ResembleAI/Dramabox).

It includes **only the audio-generation weights**:

| Component | Keys prefix | Description |
|---|---|---|
| Audio VAE | `audio_vae.*` | Encoder / decoder VAE operating on mel-spectrograms (BF16) |
| Vocoder | `vocoder.vocoder.*` | HiFi-GAN style neural vocoder (BF16) |
| BWE Generator | `vocoder.bwe_generator.*` | Bandwidth extension generator (BF16) |
| Mel STFT | `vocoder.mel_stft.*` | Mel filterbank + STFT forward/inverse basis (BF16) |

All weights are stored in **BFloat16**.

## File

| File | Contents |
|---|---|
| `dramabox-audiovae-vocoder.safetensors` | audio_vae + vocoder (merged) |

## Usage

```python
from safetensors import safe_open

tensors = {}
with safe_open("dramabox-audiovae-vocoder.safetensors", framework="pt", device="cpu") as f:
    for key in f.keys():
        tensors[key] = f.get_tensor(key)

print(list(tensors.keys())[:5])
```

## Source

Extracted from the original
[ResembleAI/Dramabox](https://huggingface.co/ResembleAI/Dramabox) checkpoint.
Please refer to the original repository for licensing details.