Where the words go
A browser page can generate speech in more than one way. A cloud service receives your text and returns audio. The browser SpeechSynthesis API can use either a local operating-system voice or a remote voice, depending on the voice and browser. A local neural model downloads model files, then generates speech on your device.
TTS Lines uses Supertonic 3 for most languages and a separate BlueTTS model with ReNikud for Hebrew. Your script text and generated audio are processed in the browser. The first model download comes from a public model host; subsequent availability depends on your browser cache. The site does not need an account to use the editor.
Privacy has practical limits
A local workflow avoids sending the script to a speech API, but it does not make every network request disappear. Loading the website, fonts, and model can still involve normal web requests. Browser storage can also be cleared by the user or by the browser. Keep your own copy of a valuable script and download the files you need.
This matters most when you are working with unpublished material. Verify the voice engine shown in the product before assuming text stays local, and check whether a tool's export is generated by a server.
What about other languages?
Local model support is tied to each model's trained languages. Supertonic 3 has ten fixed voice styles across 31 languages, including English, Spanish, Arabic, Hindi, Japanese, and Korean. Hebrew uses a different local model with four voice presets. Mandarin is not offered for export.