Fact · Speech recognition
Google said in 2019 that its new on-device recognizer design offered the same accuracy as its traditional server-based models in a much smaller model.
Checked against its source on October 16, 2026
In the source's own words
“The RNN-T we trained offers the same accuracy as the traditional server-based models but is only 450MB”
An All-Neural On-Device Speech Recognizer, Johan Schalkwyk, Google Research Blog, 2019, Offline Recognition section.
Open the original- What it measures
- Google then compressed the model to 80 MB so it could run on phones.
- What it does not show
- Google's own comparison on its own tests, for American English in 2019. Accuracy of on-device and cloud tools still varies by language and device.
Source
Cited on
- Speech Recognition: What It Is and How Well It WorksTopic
- What Is Dictation Software? Types and How It WorksGuide
- What Is Dictation? Voice Typing and Talk to Text ExplainedGuide
- On-Device vs Cloud Speech Recognition: The TradeoffsGuide
- Offline Voice Typing on iPhone and AndroidGuide
- History of Speech Recognition: From Audrey to WhisperGuide
- Is offline speech recognition more private than cloud speech recognition?Question
- Is on-device speech recognition as accurate as cloud?Question
- Cloud dictationGlossary