Ara Studio optional local speech generation The inference implementation follows Supertone's MIT-licensed Web example: Copyright (c) 2025 Supertone Inc. See SUPERTONIC-MIT.txt. https://github.com/supertone-oss-archive/supertonic/blob/main/web/helper.js ONNX Runtime Web 1.31.1, Microsoft Corporation, MIT licence. See ONNXRUNTIME-LICENSE.txt and ONNXRUNTIME-NOTICES.txt. https://github.com/microsoft/onnxruntime/tree/v1.31.1 The optional Supertonic 2 model and voice styles are downloaded only when Generate speech is chosen. They are not bundled with Ara Studio. Model provider: Supertone Inc. Model licence: OpenRAIL-M. See MODEL-LICENSE.txt, including use restrictions. Pinned model revision: 75e6727618a02f323c720cba9478152d4bc16ca4 https://huggingface.co/Supertone/supertonic-2/tree/75e6727618a02f323c720cba9478152d4bc16ca4 Model hosting can be changed with VITE_STUDIO_TTS_MODEL_URL at build time. The selected host receives model-file requests. Dialogue inference runs in a browser worker on the user's device; Ara does not upload the script to it. Native device read-through uses the browser/OS voice service separately.