Speech is generated on your Mac without an Ollama server.
Version 1.0.4: removes local build paths from the application and speech provider executables. Speech and voice data are unchanged.
Version 1.0.3: moves downloads, source code, support and signed updates to GitHub. Existing installations can update normally through the original server. Speech, voices and saved settings are unchanged.
Version 1.0.2: makes all 12 English character voices available in VoiceOver's English (United States) voice list. They remain grouped as English Character Voices in this app. Existing voice names and selections are unchanged.
Version 1.0.1: fixes installation failures caused by accented voice filenames. Displayed voice names and saved voice selections are unchanged. If macOS refuses to open version 1.0.0, download the current disk image and copy the app into Applications. There is no need to change Gatekeeper or other security settings.
Version 1.0.0: native, offline VoiceOver speech in five languages with 102 voices; streaming rate, pitch and volume controls; voice preview and signed updates.
Copy Phonon for Mac from the disk image into Applications, open it and wait for Ready. Choose a language and voice, then use Test Voice. Refresh Voices updates the system list; it may take a short time to appear.
In VoiceOver Utility, add a Phonon voice to your speech voices and select it. Keep a familiar Apple voice available as an alternative. Closing this application does not disable the voices.
The first system utterance can take several seconds while Core ML prepares the engine. Subsequent utterances stream without waiting for the whole text.
VoiceOver supplies rate, pitch and volume. Phonon generates streaming audio; Sonic adjusts rate and pitch during generation. Navigating cancels obsolete speech.
English, French, German, Spanish and Portuguese. Voices are grouped by their published locale. English character voices have their own group in this app and appear under English (United States) in VoiceOver. They speak English; this grouping does not claim that every character has a US accent.
Check for updates offers Never, Hourly and Daily. Daily is the default. Checks run when this settings app opens and at the selected interval while it stays open; no login item or update helper runs after you quit. Check Now performs a manual check.
Updates require a valid Phonon signature, matching download size and hash, the correct Apple developer signature and notarization. Installation closes and reopens this app, preserving its preferences and prepared model data. VoiceOver can continue speaking through its loaded voice; restart VoiceOver when convenient after an update. Do not interrupt installation.
Speech stays on this Mac. No spoken text or audio is saved by the app. The settings app contacts GitHub to check for updates and download an update you approve. It sends no speech text. Automatic checks can be disabled.
Application by Andre Louis. Native engine and voice catalogue by Gradium, building on Kyutai's Pocket TTS and Mimi. Streaming rate and pitch processing uses Sonic by Bill Cox.
Development assistance from OpenAI Codex. The original third-party licence texts are included in the application's Licenses resource folder.
The multilingual 40M model is licensed under CC BY-NC 4.0, which permits noncommercial sharing subject to its terms. This app is a free, independent accessibility project, not an official or endorsed Gradium product.
The original checkpoint was converted to Core ML using Gradium's export tool; the model was not retrained. The runtime is MIT OR Apache 2.0, and Gradium's published voice catalogue and Sonic are Apache 2.0. Their original licence texts, source links and notices are included in the application's Licenses folder.
Application source code is GPL-3.0-or-later; third-party code and model assets retain the separate licences described above.