Google Launches Offline On-Device Speech Recognition and TTS APIs
In short: Google has expanded its AI Edge Gallery with new on-device speech capabilities, enabling offline, low-latency transcription and synthesis. Developers can now integrate these high-performance models via Android's ML Kit or cross-platform MediaPipe Tasks, supporting over 100 languages with significant energy efficiency.
This summary was generated automatically by AI from Google Developers's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1ML Kit now includes production-grade GenAI Speech Recognition and TTS APIs for Android.
- 2MediaPipe Tasks provides turnkey ASR and TTS support for web, desktop, and mobile.
- 3New lightweight models (Parakeet-TDT, PocketTTS, Kokoro-82M, Phonon-31M) added as .litertlm bundles.
- 4Offline bilingual code-switching support for mixed-language TTS scenarios.
| Parameter | Before | Now |
|---|---|---|
| ASR Speed | Not specified | 10x real-time |
| TTS Time-to-First-Audio | Not specified | ~100 ms |
| TTS Synthesis Speed | Not specified | Up to 50x real-time |
Why it matters
These tools allow building voice-first interfaces that function entirely offline on mobile devices, removing network dependency for conversational agents while maintaining sub-100ms response latency.
What it means for AI agents and contact centers
This is a major opportunity to migrate from cloud-dependent ASR/TTS to local, hardware-accelerated pipelines for voice agents. Compare the 8-9% Word Error Rate (FLEURS) of these local models against your current cloud-based solutions to see if local execution can maintain sufficient accuracy while eliminating API costs and connectivity risks for your clients.
🧪 Worth Testing
The ability to run multilingual ASR and expressive TTS locally without APK bloat is a significant advancement for mobile voice-AI agents.
Sources
- Google DevelopersOfficialPrimary sourceOriginal article →„On-Device Speech Recognition and Text-to-Speech in Google AI Edge Gallery“10 Nov 2026, 02:00
- Published by source
- 10 Nov 2026, 02:00
- Found by our system
- 10 Nov 2026, 19:24
- Summary generated
- 10 Nov 2026, 21:26
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy