Google introduced on-device Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) in Google AI Edge Gallery on 10 November, letting phones and wearables understand spoken language and speak back entirely offline. The lightweight speech models pair with on-device small language models such as Gemini Nano and Gemma 4, so apps can stream live transcriptions as you speak and synthesise natural spoken replies locally โ€” with no network connection needed after the initial model download.

On Android, the new capabilities arrive through ML Kit's GenAI Speech Recognition and Text-to-Speech APIs. Google says its system speech models transcribe across 100+ languages at 10x real-time speed with an 8.0% to 9.0% word error rate on the FLEURS benchmark, while the few-step diffusion TTS model starts speaking in about 100 milliseconds and synthesises 24 kHz audio at up to 50x real-time speed. Both models are integrated into the operating system, so Google says they add 0 MB to an app's APK size โ€” that claim applies to the Android path only. These are Google's own figures, not independently measured.

For cross-platform apps, Google optimised new open-weight speech bundles for MediaPipe Tasks and LiteRT โ€” Parakeet-TDT 0.6B for recognition, and PocketTTS 100M, Kokoro-82M and Phonon 31M for speech generation โ€” each shipping as a single .litertlm bundle that runs on iOS, macOS, Windows, Linux and the web as well as Android.

Google is demonstrating the local voice pipeline with two showcases inside AI Edge Gallery: an XR Language Teacher that combines camera vision, speech and an on-device model for multimodal voice tutoring with no connection at all, and voice-enabled AI Chat, giving the app's main chat feature a spoken interface. The latest AI Edge Gallery app is available on Android and iOS.

Sources