{"id":90575,"date":"2026-05-15T16:34:01","date_gmt":"2026-05-15T16:34:01","guid":{"rendered":"https:\/\/youzum.net\/supertone-releases-supertonic-v3-on-device-text-to-speech-model-with-31-language-support-fewer-reading-failures-and-expression-tags\/"},"modified":"2026-05-15T16:34:01","modified_gmt":"2026-05-15T16:34:01","slug":"supertone-releases-supertonic-v3-on-device-text-to-speech-model-with-31-language-support-fewer-reading-failures-and-expression-tags","status":"publish","type":"post","link":"https:\/\/youzum.net\/es\/supertone-releases-supertonic-v3-on-device-text-to-speech-model-with-31-language-support-fewer-reading-failures-and-expression-tags\/","title":{"rendered":"Supertone Releases Supertonic v3: On-Device Text-to-Speech Model with 31-Language Support, Fewer Reading Failures, and Expression Tags"},"content":{"rendered":"

Supertone released Supertonic 3, the third generation of its on-device, ONNX-based text-to-speech system. Supertonic 3 ships with 31-language support, improved reading accuracy, fewer repeat and skip failures, and v2-compatible public ONNX assets. It is Lightning Fast, On-Device, Multilingual and Accurate TTS.<\/p>\n

What Changed from v2 to v3<\/strong><\/h2>\n

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. Version 2 supported English, Korean, Spanish, Portuguese, and French. Version 3 adds Japanese, Arabic, Bulgarian, Czech, Danish, German, Greek, Estonian, Finnish, Croatian, Hungarian, Indonesian, Italian, Lithuanian, Latvian, Dutch, Polish, Romanian, Russian, Slovak, Slovenian, Swedish, Turkish, Ukrainian, and Vietnamese \u2014 31 total ISO language codes. There is also a special na<\/code> fallback for text whose language is unknown or outside the supported set.<\/p>\n

The model grows modestly to accommodate the added languages. 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. The update also brings the total disk footprint of the public ONNX assets to 404 MB<\/strong>. Additionally, Supertone recently launched the Voice Builder<\/a><\/strong>, allowing developers to create custom, edge-native TTS models from their own voice recordings.<\/p>\n

Expressive Tags<\/strong><\/h2>\n

One new capability in v3 that wasn\u2019t present in v2 is expressive tag support. Supertonic 3 supports simple expression tags such as <laugh><\/code>, <breath><\/code>, and <sigh><\/code>. These let you embed prosodic cues directly into input text without a separate preprocessing step or a separate model for expressiveness. For engineers building voice interfaces or accessibility tools, this means you can specify breathing pauses or laughter inline in your text payload.<\/p>\n

Architecture and Runtime<\/strong><\/h2>\n

The underlying architecture carries over from prior versions: a speech autoencoder that encodes waveforms into continuous latent representations, a flow-matching based text-to-latent module that maps text to audio features, and a duration predictor that controls natural timing. Flow matching is a generative modeling technique that learns a vector field to transform a simple distribution into a target distribution \u2014 it samples faster than diffusion models at low step counts, which is why Supertonic can produce usable output in just 2 inference steps. To further refine output, v3 integrates Length-Aware Rotary Position Embedding (LARoPE)<\/strong> for superior text-speech alignment and utilizes a Self-Purifying Flow Matching<\/strong> technique during training to remain robust against noisy data labels.<\/p>\n

On runtime efficiency, Supertonic 3 runs fast on CPU, even compared with larger baselines measured on A100 GPU, and uses substantially less memory. It does not require a GPU, which makes local, browser, and edge deployment much easier. <\/p>\n

Reading Accuracy<\/strong><\/h2>\n

Across measured languages, 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. WER (Word Error Rate) and CER (Character Error Rate) are standard TTS readability metrics: you synthesize a passage, run ASR over the output, and compare the transcription to the original text. CER is used for languages without clear word boundaries; the others use WER. The system\u2019s efficiency is best demonstrated on extreme edge hardware; it achieves an average RTF of 0.3x<\/strong> on an Onyx Boox Go 6<\/strong> (an E-ink e-reader) in airplane mode. Furthermore, the ecosystem has expanded to include Flutter (with macOS support)<\/strong>, .NET 9<\/strong>, and Go<\/strong>, while the web implementation leverages onnxruntime-web<\/code> for pure client-side execution.<\/p>\n

Text Normalization<\/strong><\/h2>\n

A differentiating property carried forward from v2 is built-in text normalization. Supertonic handles complex surface forms \u2014 financial expressions like $5.2M<\/code>, phone numbers with area codes and extensions like (212) 555-0142 ext. 402<\/code>, time and date formats like 4:45 PM on Wed, Apr 3, 2024<\/code>, and technical units like 2.3h<\/code> and 30kph<\/code> \u2014 without any preprocessing pipeline or phonetic annotations. The financial expression \u201c$5.2M\u201d must read as \u201cfive point two million dollars,\u201d and \u201c$450K\u201d as \u201cfour hundred fifty thousand dollars.\u201d All four competing systems failed this. The technical unit \u201c2.3h\u201d must read as \u201ctwo point three hours\u201d and \u201c30kph\u201d as \u201cthirty kilometers per hour.\u201d All four competitors also failed this category. The competing systems evaluated include ElevenLabs Flash v2.5, OpenAI TTS-1, Gemini 2.5 Flash TTS, and Microsoft.<\/p>\n

\n
\"\"
https:\/\/github.com\/supertone-inc\/supertonic<\/figcaption><\/figure>\n<\/div>\n

Getting Started<\/strong><\/h2>\n

The Python SDK install is pip install supertonic<\/code>. On first run, the SDK downloads the model assets from Hugging Face automatically. A minimal example:<\/p>\n

\n
\n
\n
\n
<\/div>\n
<\/div>\n
<\/div>\n<\/div>\n
Copy Code<\/span>Copied<\/span>Use a different Browser<\/span><\/a><\/div>\n<\/div>\n
from supertonic import TTS\ntts = TTS(auto_download=True)\nstyle = tts.get_voice_style(voice_name=\"M1\")\ntext = \"A gentle breeze moved through the open window while everyone listened to the story.\"\nwav, duration = tts.synthesize(text, voice_style=style, lang=\"en\")\ntts.save_audio(wav, \"output.wav\")\nprint(f\"Generated {duration:.2f}s of audio\")<\/code><\/pre>\n<\/div>\n<\/div>\n
Marktechpost\u2019s Visual Explainer<\/strong><\/h2>\n
\n
  <\/p>\n
\n
\n
\n        <\/span><\/span><\/span><\/span>\n      <\/div>\n
      Supertonic 3 \u2014 Developer Guide<\/span>\n    <\/p><\/div>\n
    1 \/ 7<\/span>\n  <\/p><\/div>\n
  <\/p>\n
\n
<\/div>\n<\/div>\n
  <\/p>\n
\n
\n
      <\/p>\n
\n        Overview<\/span>\n
Supertonic 3: On-Device TTS,
Now in 31 Languages<\/h2>\n
Supertonic 3 is a lightweight, open-weight text-to-speech system by Supertone Inc. It runs entirely via ONNX Runtime on your device \u2014 no cloud, no API call, no data leaving your machine. v3 expands from 5 to 31 languages, adds expressive tags, reduces reading failures, and stays compatible with the v2 ONNX interface.<\/p>\n
\n
\n            31<\/span>
\n            Languages<\/span>\n          <\/div>\n
\n            ~99M<\/span>
\n            Parameters<\/span>\n          <\/div>\n
\n            404 MB<\/span>
\n            ONNX Assets<\/span>\n          <\/div>\n
\n            MIT<\/span>
\n            Code License<\/span>\n          <\/div>\n<\/div>\n<\/div>\n
      <\/p>\n
\n        What\u2019s New in v3<\/span>\n
Four Core Improvements Over Supertonic 2<\/h2>\n
Version 3 is a focused upgrade \u2014 same inference contract, meaningfully better output.<\/p>\n