React Native Executorch Supertonic

--- license: openrail language: - en - es - fr - pl - el - id - zh - ja - ru - tr - da - et - hr - pt - vi - sl - cs - it - uk - ko - sv - lv - hu - fi - ar - nl - sk - bg - hi - de - lt base_model: - Supertone/supertonic-3 pipeline_tag: text-to-speech library_name: executorch --- # supertonic This repository hosts the **supertonic** models exported for the [React Native ExecuTorch](https://www.npmjs.com/package/react-native-executorch) library as ExecuTorch `.pte` programs, ready to run on device. ## Variants | Path | Component | Backend | Precision | | --- | --- | --- | --- | | `mlx/duration_predictor_mlx_fp32.pte` | duration_predictor | mlx | fp32 | | `mlx/text_encoder_mlx_fp32.pte` | text_encoder | mlx | fp32 | | `mlx/vector_estimator_mlx_fp32.pte` | vector_estimator | mlx | fp32 | | `mlx/vocoder_mlx_fp32.pte` | vocoder | mlx | fp32 | | `vulkan/duration_predictor_vulkan_fp16.pte` | duration_predictor | vulkan | fp16 | | `vulkan/text_encoder_vulkan_fp16.pte` | text_encoder | vulkan | fp16 | | `vulkan/vector_estimator_vulkan_fp16.pte` | vector_estimator | vulkan | fp16 | | `vulkan/vocoder_vulkan_fp16.pte` | vocoder | vulkan | fp16 | | `xnnpack/duration_predictor_xnnpack_fp32.pte` | duration_predictor | xnnpack | fp32 | | `xnnpack/text_encoder_xnnpack_fp32.pte` | text_encoder | xnnpack | fp32 | | `xnnpack/vector_estimator_xnnpack_fp32.pte` | vector_estimator | xnnpack | fp32 | | `xnnpack/vector_estimator_xnnpack_int8.pte` | vector_estimator | xnnpack | int8 | | `xnnpack/vocoder_xnnpack_fp32.pte` | vocoder | xnnpack | fp32 | | `xnnpack/vocoder_xnnpack_int8.pte` | vocoder | xnnpack | int8 | ## XNNPACK int8 `xnnpack/vector_estimator_xnnpack_int8.pte` and `xnnpack/vocoder_xnnpack_int8.pte` are 8da8w (int8 dynamic activations, int8 per-channel weights) on the Linear layers, with the ConvNeXt pointwise convs moved onto Linears. The int8 set reuses the fp32 duration predictor and text encoder. LibriSpeech test-clean texts, voice F1, 8 steps, WER of a whisper base.en round trip: | | WER (57 short) | WER (8 long) | | --- | --- | --- | | fp32 | 3.36 | 3.32 | | fp32, other noise seed | 3.26 | 3.80 | | int8 | 3.74 | 3.01 | ## Unshipped backends A backend directory carrying a `NOTES.md` and no `.pte` was exported and not published. Its note records what went wrong. - `coreml` ([why](https://huggingface.co/software-mansion/react-native-executorch-supertonic/...

--- license: openrail language: - en - es - fr - pl - el - id - zh - ja - ru - tr - da - et - hr - pt - vi - sl - cs - it - uk - ko - sv - lv - hu - fi - ar - nl - sk - bg - hi - de - lt base_model: - Supertone/supertonic-3 pipeline_tag: text-to-speech --- # Supertonic 3 TTS This repository hosts the **Supertonic 3** model for the [React Native ExecuTorch](https://github.com/software-mansion/react-native-executorch) library. It performs text-to-speech synthesis supporting **30+ languages**, with a single voice style per language. The model is composed of four sub-models that run sequentially: 1. **Duration predictor** — estimates speech duration from text 2. **Text encoder** — encodes text into a style-conditioned representation 3. **Vector estimator** — flow-matching denoiser that generates the audio latent 4. **Vocoder** — decodes the latent into a 44.1 kHz waveform ## Compatibility These models were exported using **v1.3.1** of ExecuTorch and no forward compatibility is guaranteed. Older versions of the runtime may not work with these files. The models are intended to be used within the React Native ExecuTorch package. If you want to use them outside the package, make sure your runtime is compatible with the ExecuTorch version used to export the `.pte` files and follow the example scripts to run the models. ## Backends | Backend | Description | RTF (Apple Silicon) | |---------|-------------|---------------------| | `xnnpack` | CPU-optimized via XNNPACK delegate | ~0.07 (14× faster than real-time) | | `mlx` | Apple Silicon GPU via MLX delegate | ~0.026 (38× faster than real-time) | ## Repository Structure ``` . ├── config.json # Backend-agnostic model manifest ├── unicode_indexer.json # Character-to-id mapping for text preprocessing ├── voices/ # Pre-computed speaker embeddings │ ├── M1.json │ ├── M2.json │ ├── F1.json │ └── ... ├── xnnpack/ # XNNPACK-exported .pte files └── mlx/ # MLX-exported .pte files ``` Each `.pte` file exposes two methods: - `forward` — the sub-model inference - `get_dynamic_dims_forward` — returns per-input `[rank, 3]` shape constraints (`[min, max, step]`) for runtime input validation

Source context: 4814 downloads · 1 likes · Pipeline text-to-speech · Library executorch · Repo software-mansion/react-native-executorch-supertonic