wtf is qwen3-tts-android?

Qwen3 TTS Android is an Android app that turns text into spoken audio entirely on the phone itself, without sending anything to a server. It runs the Qwen3-TTS speech model locally through a native runtime called qwen3-tts.cpp, and saves the generated audio files directly on the device. The app is built with Jetpack Compose and organizes its features into four screens: a Studio screen for generating speech, a Voices screen for managing voice profiles, a History screen showing past generations with playback, delete, and export to WAV file options, and a Settings screen. One of the more distinctive features is the ability to create a custom voice profile by recording your own voice with the microphone, which the app then uses to extract a speaker embedding, a compact representation of how that voice sounds, so generated speech can match it. On first launch, the app downloads a specific quantized model package, made up of a tokenizer file and a talker model file, from a public model repository on Hugging Face, storing them privately within the app. The app shows generation progress including elapsed time and an estimated audio length based on past history, and it currently runs only on the CPU rather than requiring a specialized processor. Building the app from source requires Android Studio or the Android SDK, the Android NDK, JDK 17, and an arm64 Android device running Android 12 or newer, since the native audio engine is compiled specifically for that processor architecture. The build process involves initializing git submodules before compiling either a debug or release APK with Gradle. The project is released under the MIT license.