---
license: apache-2.0
pipeline_tag: text-to-speech
library_name: rlx
tags:
- zipvoice
- onnx
- tts
- rlx
---

# ZipVoice ONNX (RLX staging)

ZipVoice ONNX text encoder + flow decoder for RLX.

| Field | Value |
|---|---|
| **Hub id** | [`eugenehp/zipvoice`](https://huggingface.co/eugenehp/zipvoice) |
| **Kind** | Converted / re-laid-out for RLX from an upstream checkpoint. |
| **RLX crate** | [`rlx-zipvoice`](https://github.com/MIT-RLX/rlx-models/tree/main/crates/rlx-zipvoice) |
| **Upstream** | https://huggingface.co/k2-fsa/ZipVoice |

## Quick start

```bash
hf download eugenehp/zipvoice --local-dir .
cargo run -p rlx-zipvoice --release -- --model-dir .
```

## File highlights

- `fm_decoder.onnx` (455.3 MiB)
- `zipvoice.f16.gguf` (267.9 MiB)
- `onnx/vocoder_spec.onnx` (51.6 MiB)
- `text_encoder.onnx` (16.8 MiB)
- `encoder_body.onnx` (16.8 MiB)
- `tokens.txt` (2.5 KiB)

## Do not use as the main runtime pack

These files may be present for historical / staging reasons but are **not** the supported RLX load path:

- `zipvoice.f16.gguf`

Community `*.f16.gguf` / LM-only Q4 packs are format wraps — they are not drop-in replacements for the RLX primary file above.

## Run with RLX

Clone [rlx-models](https://github.com/MIT-RLX/rlx-models), place this repo under `weights/tts/zipvoice` (or pass the path explicitly), then:

```bash
cargo run -p rlx-zipvoice --release -- --model-dir .
```

## License

Apache License 2.0 — see `LICENSE`. Inherit upstream terms when redistributing.

Original weights and authorship: https://huggingface.co/k2-fsa/ZipVoice

## Maintenance

Cards and LFS attrs are regenerated from the local `weights/` tree in [rlx-models](https://github.com/MIT-RLX/rlx-models) via `python3 scripts/prepare_weights_hf.py`.