Applio for Mac v3.6.4.2
The big topic in this release is accessibility. The app now works properly with VoiceOver, and there is nothing to set up or enable.
Turn VoiceOver on and Applio speaks. Every job announces when it starts, how far along it is, and when it finishes or fails: conversions (single and batch), training with its preprocessing and feature extraction, text to speech, model downloads, voice blending, plugin installs and realtime conversion. The Active Processes menu lists live jobs with their progress, and the Process Dashboard (Cmd+Shift+P) reads out live numbers for each run: current epoch, best loss, files converted, estimated time left. Every path field has a Browse button that opens a proper macOS picker, and the file dropdowns show readable names instead of long file paths. Speech only runs whilst VoiceOver is on. Everyone gets the completion and failure toasts, and a sound cue when a job ends.
All of this was contributed back to the main Applio project. Eight pull requests, all merged, so people using the standard version benefit too.
Quitting is also safer now. Closing the window or pressing Cmd+Q during a conversion asks first instead of pulling the app down mid-job, and a confirmed quit cancels the work cleanly.
From upstream Applio since the last release:
- multi-speaker training
- an rmvpe option for high-register voices, in the training settings
- training runs faster: TF32 is used on Apple Silicon, and discriminator gradients are skipped during generator loss evaluation
- fixes for model information output, model blending and TensorBoard links
- the translation pass is complete: the strings that used to stay English are translated, and the language picker shows each language in its own spelling
As before, the app is signed and notarized. Download the DMG, drag Applio to your Applications folder, and your existing data and settings are left alone. The first launch downloads the model files, about 2 GB.
Full changelog: CHANGELOG.md in the repository.