Audio engines / Motores de audio

These tools run on your device. No paid inference service is used. Retain the licenses and source notices when redistributing the engines.

• FFmpeg.wasm wrapper (MIT), core 0.12.10 (GPL-2.0-or-later). Wrapper source · Core source and build instructions · FFmpeg source.
• Transformers.js 4.3.0 (Apache-2.0). Source.
• ONNX Runtime Web (MIT). Source.
• English phonemizer wrapper (Apache-2.0), bundled eSpeak NG engine (GPL-3.0-or-later). Wrapper source · Engine source.
• Spanish eSpeak NG Emscripten engine (GPL-3.0). Distributed unmodified. License · Port source · Engine source and build instructions.
• Demucs Web 1.0.2 (MIT), Meta Demucs (MIT). Port source · Original source · Model.
• Kokoro-82M (Apache-2.0). Original model · ONNX weights and voices.
• Whisper (MIT). Source · Multilingual ONNX model.
• RNNoise (BSD-3-Clause, Xiph.Org), through FFmpeg's arnndn filter; model «somnolent-hogwash» from rnnoise-models (not subject to copyright according to its authors). Algorithm source · Model.
• Chatterbox (MIT). Source · English ONNX model.

Local integration sources are included with generated engine assets. Upstream licenses, notices and build recipes apply to upstream components.