supertone-inc/supertonic

Supertonic — Lightning Fast, On-Device, Accurate TTS

Supertonic is a lightning-fast, on-device multilingual text-to-speech system designed for local inference with minimal overhead. Powered by ONNX Runtime, it runs entirely on your device—no cloud, no API calls, no privacy concerns.

✨ Highlights

• ⚡ Blazingly Fast — Low-latency, real-time synthesis across desktop, browser, mobile, and edge — fast enough to turn an entire webpage into audio in under a second
• 🌍 31-Language Multilingual — Synthesize directly from text across 31 languages, or pass lang="na" to let Supertonic process the text language-agnostically when you don’t know the input language — no separate language adapters needed
• 🪶 99M-Parameter Open-Weight Model — A compact, fully open-weight checkpoint — a fraction of the size of 0.7B–2B class open TTS systems — for smaller downloads, faster cold starts, and lower memory footprint
• 📱 Edge-Device Ready — Runs locally on desktop, mobile, browsers, and resource-constrained hardware like Raspberry Pi or e-readers, with zero network dependency, complete privacy, and no GPU required
• 🔊 44.1kHz High-Quality Audio — Outputs studio-grade 44.1kHz 16-bit WAV directly, ready for production playback without any external upsampler
• 🎭 Expression Tags — 10 inline tags (e.g. <laugh>, <breath>, <sigh>) bring natural human nuance into generated speech without prompt engineering or reference audio
• 🛠️ Multi-Runtime SDKs — Ready-to-use examples through ONNX Runtime across Python, Node.js, Browser (WebGPU), Java, C++, C#, Go, Swift, iOS, Rust, and Flutter

🌍 Supported Languages (31)

Arabic (ar), Bulgarian (bg), Croatian (hr), Czech (cs), Danish (da), Dutch (nl), English (en), Estonian (et), Finnish (fi), French (fr), German (de), Greek (el), Hindi (hi), Hungarian (hu), Indonesian (id), Italian (it), Japanese (ja), Korean (ko), Latvian (lv), Lithuanian (lt), Polish (pl), Portuguese (pt), Romanian (ro), Russian (ru), Slovak (sk), Slovenian (sl), Spanish (es), Swedish (sv), Turkish (tr), Ukrainian (uk), Vietnamese (vi)

Not sure which language your text is in? Pass lang="na" and Supertonic will handle the input in a language-agnostic way — no explicit language tag required.

📰 Update News

• 2026.05.18 - Python SDK v1.3.1 adds supertonic serve, a local HTTP server with native /v1/tts and OpenAI-compatible /v1/audio/speech endpoints. See the serve documentation.
• 2026.05.18 - Voice Builder now supports Supertonic 3. Create a permanent custom voice profile for Supertonic and download version-specific JSON files for both Supertonic 2 and Supertonic 3. If you already created a Supertonic 2 voice, the matching Supertonic 3 JSON is now available from My Page.
• 2026.04.29 - 🎉 Supertonic 3 released with 31-language support, improved reading accuracy, fewer repeat/skip failures, and v2-compatible public ONNX assets. Demo | Models
• 2026.01.22 - Voice Builder is now live! Turn your voice into a deployable, edge-native TTS with permanent ownership.
• 2026.01.06 - 🎉 Supertonic 2 released with 5-language support. The v2 code path is preserved on the release/supertonic-2 branch.
• 2025.12.10 - Added supertonic PyPI package! Install via pip install supertonic. For details, visit supertonic-py documentation
• 2025.12.10 - Added 6 new voice styles (M3, M4, M5, F3, F4, F5). See Voices for details
• 2025.12.08 - Optimized ONNX models via OnnxSlim now available on Hugging Face Models
• 2025.11.24 - Added Flutter SDK support with macOS compatibility

Quick Start

Install the Python SDK and generate speech immediately. On the first run, Supertonic downloads the model assets from Hugging Face automatically.

1

pip install supertonic

Python

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

from supertonic import TTS

# First run downloads the model from Hugging Face automatically.
tts = TTS(auto_download=True)

style = tts.get_voice_style(voice_name="M1")

text = "Supertonic is a lightning fast, on-device TTS system."

wav, duration = tts.synthesize(
    text=text,
    lang="en",                      # Language code (e.g., "en", "ko", "na" for language-agnostic)
    voice_style=style,              # Voice style object
    total_steps=8,                  # Quality: 5 (low) to 12 (high), default 8 (medium)
    speed=1.05,                     # Speed: 0.7 (slow) to 2.0 (fast)
)
# wav: numpy array of shape (1, num_samples,) with dtype=np.float32, sampled at 44100 Hz
# duration: numpy array of shape (1,) containing the duration of the generated audio in seconds

tts.save_audio(wav, "output.wav")
# import soundfile as sf
# sf.write("output.wav", wav.squeeze(), 44100)

print(f"Generated {duration[0]:.2f}s of audio")

Local HTTP Server

The Python SDK can also run Supertonic as a local HTTP service. This is useful when you want to call Supertonic from tools that already speak HTTP, such as local agents, browser extensions, Electron apps, workflow automation tools, or OpenAI-compatible audio clients.

1
2

pip install 'supertonic[serve]'
supertonic serve --host 127.0.0.1 --port 7788

Once running, use the native POST /v1/tts endpoint or the OpenAI-compatible POST /v1/audio/speech endpoint. The server also exposes interactive OpenAPI docs at http://127.0.0.1:7788/docs. See the supertonic-py serve guide for request examples, batch synthesis, and custom Voice Builder JSON import.

Getting Started

First, clone the repository:

1
2

git clone https://github.com/supertone-inc/supertonic.git
cd supertonic

Prerequisites

Before running the examples, download the ONNX models and preset voices, and place them in the assets directory:

Note: The Hugging Face repository uses Git LFS. Please ensure Git LFS is installed and initialized before cloning or pulling large model files.

• macOS: brew install git-lfs && git lfs install
• Generic: see https://git-lfs.com for installers

1
2

git lfs install
git clone https://huggingface.co/Supertone/supertonic-3 assets

Some language examples need native runtimes:

• Go: install the ONNX Runtime C library. On macOS, brew install onnxruntime is enough; the Go example auto-detects Homebrew paths.
• Java: use a JDK, not just a JRE. On macOS, brew install openjdk@17 works.
• C#: targets .NET 9 and allows major-version roll-forward, so .NET 9 or newer runtimes can run it.

Then run the Python example:

1
2
3

cd py
uv sync
uv run example_onnx.py

This generates outputs/output.wav using the default preset voice.

Other Runtime Examples

1
2
3

cd nodejs
npm install
npm start

1
2
3

cd web
npm install
npm run dev

1
2
3

cd java
mvn clean install
mvn exec:java

1
2
3
4

cd cpp
mkdir build && cd build
cmake .. && cmake --build . --config Release
./example_onnx

1
2
3

cd csharp
dotnet restore
dotnet run

1
2
3

cd go
go mod download
go run example_onnx.go helper.go

1
2
3

cd swift
swift build -c release
.build/release/example_onnx

1
2
3

cd rust
cargo build --release
./target/release/example_onnx

1
2
3

cd ios/ExampleiOSApp
xcodegen generate
open ExampleiOSApp.xcodeproj

Technical Details

• Runtime: ONNX Runtime for cross-platform inference
• Browser Support: onnxruntime-web for client-side inference
• Batch Processing: Supports batch inference for improved throughput
• Audio Output: Outputs 44.1kHz 16-bit WAV files

Performance Highlights

Supertonic 3 is designed for practical on-device inference: compact enough to run locally, while staying competitive with much larger open TTS systems.

Reading Accuracy

Evaluated on the Minimax-MLS-test benchmark, Supertonic 3 stays within a competitive WER/CER range against much larger open TTS models such as VoxCPM2, while preserving a lightweight on-device deployment path. Asterisked languages (*) use CER; the others use WER.

Supertonic 2 to Supertonic 3

Compared with Supertonic 2, Supertonic 3 reduces repeat and skip failures, improves speaker similarity across the shared-language set, and expands language coverage from 5 to 31 languages. It keeps the v2-compatible public ONNX interface, so existing integrations can move to v3 with the same inference contract.

Runtime Footprint

Supertonic 3 runs fast on CPU, even compared with larger baselines measured on A100 GPU, and uses substantially less memory. The open-weight fixed-voice setting does not require a GPU, which makes local, browser, and edge deployment much easier.

Model Size

At about 99M parameters across the public ONNX assets, Supertonic 3 is much smaller than 0.7B to 2B class open TTS systems. The smaller model size is a practical advantage for download size, startup time, and on-device inference.

Voice Cloning

This open-weight repository focuses on fixed-voice, local TTS and does not include an official voice-cloning pipeline. If you want to bring your own voice to local Supertonic deployment, Voice Builder turns a short reference recording into version-specific JSON files for Supertonic 2 and Supertonic 3, so the same custom voice can move with you across supported Supertonic versions.

For a managed creation workflow, Supertone Play and the Supertone API provide hosted TTS and voice services with 700+ commercially usable preset voices. You can also listen to Supertonic 3 zero-shot samples on the official showcase.

Demo

Try it now: Experience Supertonic in your browser with our Interactive Demo, or get started with pre-trained models from Hugging Face Hub

Raspberry Pi

Watch Supertonic running on a Raspberry Pi, demonstrating on-device, real-time text-to-speech synthesis:

https://github.com/user-attachments/assets/ea66f6d6-7bc5-4308-8a88-1ce3e07400d2

E-Reader

Experience Supertonic on an Onyx Boox Go 6 e-reader in airplane mode, achieving an average RTF of 0.3× with zero network dependency:

https://github.com/user-attachments/assets/64980e58-ad91-423a-9623-78c2ffc13680

Chrome Extension

Turns any webpage into audio in under one second, delivering lightning-fast, on-device text-to-speech with zero network dependency—free, private, and effortless:

https://github.com/user-attachments/assets/cc8a45fc-5c3e-4b2c-8439-a14c3d00d91c

Programming Language Support

We provide ready-to-use TTS inference examples across multiple ecosystems:

For detailed usage instructions, please refer to the README.md in each language directory.

Natural Text Handling

Supertonic is designed to handle complex, real-world text inputs that contain natural prose, punctuation, abbreviations, and proper nouns.

🎧 View audio samples more easily: Check out our Interactive Demo for a better viewing experience of all audio examples

Overview of Test Cases:

Note: These samples demonstrate how each system handles text normalization and pronunciation of complex expressions without requiring pre-processing or phonetic annotations.

Built with Supertonic

Models & Versions

Citation

The following papers describe the core technologies used in Supertonic. If you use this system in your research or find these techniques useful, please consider citing the relevant papers:

SupertonicTTS: Main Architecture

This paper introduces the overall architecture of SupertonicTTS, including the speech autoencoder, flow-matching based text-to-latent module, and efficient design choices.

1
2
3
4
5
6
7

@article{kim2025supertonic,
  title={SupertonicTTS: Towards Highly Efficient and Streamlined Text-to-Speech System},
  author={Kim, Hyeongju and Yang, Jinhyeok and Yu, Yechan and Ji, Seunghun and Morton, Jacob and Bous, Frederik and Byun, Joon and Lee, Juheon},
  journal={arXiv preprint arXiv:2503.23108},
  year={2025},
  url={https://arxiv.org/abs/2503.23108}
}

Length-Aware RoPE: Text-Speech Alignment

This paper presents Length-Aware Rotary Position Embedding (LARoPE), which improves text-speech alignment in cross-attention mechanisms.

1
2
3
4
5
6
7

@article{kim2025larope,
  title={Length-Aware Rotary Position Embedding for Text-Speech Alignment},
  author={Kim, Hyeongju and Lee, Juheon and Yang, Jinhyeok and Morton, Jacob},
  journal={arXiv preprint arXiv:2509.11084},
  year={2025},
  url={https://arxiv.org/abs/2509.11084}
}

Self-Purifying Flow Matching: Training with Noisy Labels

This paper describes the self-purification technique for training flow matching models robustly with noisy or unreliable labels.

1
2
3
4
5
6
7

@article{kim2025spfm,
  title={Training Flow Matching Models with Reliable Labels via Self-Purification},
  author={Kim, Hyeongju and Yu, Yechan and Yi, June Young and Lee, Juheon},
  journal={arXiv preprint arXiv:2509.19091},
  year={2025},
  url={https://arxiv.org/abs/2509.19091}
}

License

This project’s sample code is released under the MIT License. - see the LICENSE for details.

The accompanying model is released under the OpenRAIL-M License. - see the LICENSE file for details.

This model was trained using PyTorch, which is licensed under the BSD 3-Clause License but is not redistributed with this project. - see the LICENSE for details.

Copyright (c) 2026 Supertone Inc.