How it works. All speech synthesis and transcription runs locally in your browser. No audio, text, or voice data is uploaded to any server. However, on first use (and after cache clears), the app downloads AI model weights (~1.2 GB, plus ~1.5 GB for voice cloning and automatic transcription) from Hugging Face (huggingface.co); browsers without WebGPU also load a runtime library from jsDelivr (cdn.jsdelivr.net). These downloads expose your IP address, browser user-agent, and referrer to those services under their respective privacy policies. Once cached, no further network requests are made during synthesis.
AI-generated content. All audio produced by this tool is synthetic speech generated by an AI model (OmniVoice). Downloaded MP3 files contain machine-readable metadata (ID3v2) identifying them as AI-generated in accordance with EU AI Act Art. 50. Do not remove or alter this metadata.
User obligations under the EU AI Act. If you publish or distribute audio generated by this tool, you are legally required to disclose its synthetic nature to your audience (Art. 50(4)), including audible disclaimers where applicable. Using this tool to impersonate real individuals without their explicit consent, to create deceptive deepfakes, or to strip or circumvent AI content-marking is prohibited and may constitute a violation of the EU AI Act and other applicable law.
No warranty. This tool is provided as-is, without warranty of any kind, for personal, creative, and research purposes. The developers accept no liability for the content users generate, for any downstream use or misuse, or for the availability, accuracy, or conduct of third-party services (Hugging Face, jsDelivr, GitHub).
Licences. The app is open source under Apache-2.0. The OmniVoice model by Xiaomi Corp. is licensed CC BY-NC (non-commercial). The voice encoder and audio decoder are Boson AI's Higgs Audio v2 tokenizer, under the Boson Higgs Audio 2 Community License; their use must follow the Llama 3 Acceptable Use Policy. Built with Higgs Materials licensed from Boson AI USA, Inc., Copyright Boson AI USA, Inc., All Rights Reserved and Meta Llama 3 licensed under the Meta Llama 3 Community License, Copyright Meta Platforms, Inc., All Rights Reserved. Transcription: Whisper by OpenAI (MIT). MP3 export: lamejs (LGPL-3.0). All licences: README.
Third-party services: Hugging Face Privacy Policy · jsDelivr Privacy Policy · GitHub Privacy Statement