Google adds offline speech recognition and text-to-speech to AI Edge Gallery

Google announced on 10 November that AI Edge Gallery now supports on-device Automatic Speech Recognition (ASR) and Text-to-Speech (TTS), paired with small on-device language models such as Gemini Nano and Gemma 4, so phones can transcribe speech live and synthesise spoken replies without any network connection. Google said voice is often the most natural way to interact with AI on phones and wearables, and advances in compact speech models now make accurate multilingual ASR and expressive TTS possible entirely on device.

Android: ML Kit covers 100+ languages and starts speaking in about 100 ms

On Android, ML Kit's latest on-device speech models transcribe more than 100 languages at 10 times real-time speed, reaching a word error rate of 8.0% to 9.0% on the FLEURS benchmark. Its text-to-speech model, powered by a compact few-step diffusion model, begins speaking in about 100 ms and synthesises 24 kHz audio at up to 50 times real-time speed, across eight expressive voices and 21 locales. Both models are integrated into the operating system, adding 0 MB to an app's APK, and benefit from hardware acceleration on NPU and TPU backends where available.

Cross-platform: open-weight models for iOS, Windows and the web

For cross-platform apps, Google has optimised new .litertlm model bundles through MediaPipe Tasks and LiteRT, including Parakeet-TDT 0.6B for speech recognition and PocketTTS 100M, Kokoro-82M and Phonon 31M for text-to-speech, running on iOS, macOS, Windows, Linux and the web alongside on-device small language models without compute contention. The Hugging Face LiteRT community also offers open-weight models such as Whisper and Qwen3-ASR.

Showcases: an offline language teacher and voice chat

Google is demonstrating two uses: the XR Language Teacher showcase, which works 100% locally on a phone, letting users point a camera at a street sign or menu, ask a question aloud and hear an explanation in their own language with authentic pronunciation in the local language; and voice-enabled AI Chat, with streaming dictation and read-aloud playback with live sentence highlighting. The features are available from today, with the AI Edge Gallery app on Android and iOS and the reference implementation open-sourced. Source: Google Developers Blog, 10 November 2026.