---
license: openrail
pipeline_tag: text-to-speech
library_name: rlx
tags:
- supertonic
- tts
- rlx
---

# Supertonic-3 (RLX staging)

Supertonic-3 ONNX + voice styles for RLX. OpenRAIL — see upstream.

| Field | Value |
|---|---|
| **Hub id** | [`eugenehp/supertonic-3`](https://huggingface.co/eugenehp/supertonic-3) |
| **Kind** | Staging redistrib of an upstream checkpoint for RLX runners. |
| **RLX crate** | [`rlx-supertonic`](https://github.com/MIT-RLX/rlx-models/tree/main/crates/rlx-supertonic) |
| **Upstream** | https://huggingface.co/Supertone/supertonic-3 |

## Quick start

```bash
hf download eugenehp/supertonic-3 --local-dir .
cargo run -p rlx-supertonic --release -- --data .
```

## File highlights

- `onnx/vector_estimator.onnx` (244.7 MiB)
- `supertonic-3.f16.gguf` (189.3 MiB)
- `onnx/vocoder.onnx` (96.7 MiB)
- `onnx/text_encoder.onnx` (34.7 MiB)
- `onnx/duration_predictor.onnx` (3.5 MiB)
- `voice_styles/F2.json` (285.6 KiB)
- `voice_styles/M2.json` (285.2 KiB)
- `voice_styles/F1.json` (285.2 KiB)
- `voice_styles/F4.json` (285.0 KiB)
- `voice_styles/M1.json` (284.9 KiB)
- `voice_styles/M4.json` (284.7 KiB)
- `voice_styles/F5.json` (284.6 KiB)
- `voice_styles/M5.json` (284.6 KiB)
- `voice_styles/F3.json` (284.0 KiB)
- `voice_styles/M3.json` (283.4 KiB)
- `onnx/unicode_indexer.json` (271.2 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:

- `supertonic-3.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/supertonic-3` (or pass the path explicitly), then:

```bash
cargo run -p rlx-supertonic --release -- --data .
```

## License

OpenRAIL — see upstream model card for use restrictions.

Original weights and authorship: https://huggingface.co/Supertone/supertonic-3

## Redistrib note

This Hub repo exists so RLX recipes have a stable fetch target. When you only need the upstream checkpoint, prefer the Upstream link above.