Phonon TTS by Gradium adds one node, PhononSpeaker, that speaks text with Phonon, Gradium's on-device text-to-speech model. The speech is generated on the player's machine. There is no server and no API key.

• speak(text) starts playing as soon as the first audio is ready, while the rest of the line is still being generated. stop() cuts it off.
• synthesize(text) generates a line ahead of time and returns it as an AudioStreamWAV.
• The speech can play through an AudioStreamPlayer, AudioStreamPlayer2D or AudioStreamPlayer3D, so a character can speak from where it stands.
• Many English voices, including character voices, come from the Phonon voice catalog.
• Loading and generation run on worker threads, so the game never waits for them.

@onready var speaker: PhononSpeaker = $PhononSpeaker

func _ready() -> void:
    speaker.loaded.connect(func(): speaker.speak("Welcome back, traveler."))
    speaker.load_model()

Models and licenses. The default model is Phonon 31m, which speaks English. It is under Apache 2.0, so you can ship it in a commercial game. The first load_model() downloads it from Hugging Face, so the game needs an internet connection once. Exported games keep it in their user:// folder. The voice catalog is under Apache 2.0 too.

The plugin can also load the Phonon 40m models, in English, French, German, Spanish and Portuguese, and Phonon 40m multilingual. These are under CC BY-NC 4.0, which does not allow commercial use. A commercial game needs a license from Gradium to use them: talk to Gradium's sales team at https://gradium.ai/contact?subject=Sales.

Platforms. Godot 4.7 or later, on macOS 15 or later (Apple silicon and Intel), Linux x86_64 and Windows x86_64. The x86_64 libraries need a CPU with AVX2. Android, iOS and web exports are not supported yet.

The plugin code is licensed under MIT OR Apache-2.0, at your option, so the Apache 2.0 license shown on this page is one of the two. The source, the full documentation and a demo project are on GitHub.