On-Device Speech Recognition and Text-to-Speech in Google AI Edge Gallery
Blog post from Hugging Face
Google AI Edge Gallery has added fully offline, on-device automatic speech recognition and text-to-speech capabilities that can work alongside small language models such as Gemini Nano and Gemma 4. The release supports Android through ML Kit’s OS-integrated speech services and cross-platform development through MediaPipe Tasks and LiteRT, with open-weight models including Parakeet-TDT, PocketTTS, Kokoro, and Phonon packaged for deployment. Demonstrations include an XR Language Teacher that combines camera input, live voice transcription, local language-model reasoning, bilingual speech output, and support for smart glasses, as well as AI Chat features for voice dictation, configurable speech settings, and sentence-synchronized read-aloud playback. ML Kit is presented as an Android-focused option offering recognition in more than 100 languages and fast, hardware-accelerated speech synthesis without increasing APK size, while MediaPipe provides higher-level APIs across major desktop, mobile, and web platforms. LiteRT supplies the underlying inference infrastructure and model bundles for developers needing greater customization, and the article directs developers to the Edge Gallery app, open-source reference code, documentation, model repositories, and device-testing resources.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| Real-time | 15 | 2,990 | 636 | 190 | -28% |
| LLM | 1 | 3,355 | 742 | 191 | -49% |
| Local AI | 1 | 82 | 54 | 19 | -51% |
| TPUs | 1 | 126 | 20 | 8 | -59% |
| Vector Search | 1 | 1,097 | 201 | 87 | -45% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.