Physix Frontier · News Briefing Card (ifanr · Oct 28, 2026)

iFlytek Input 16.0 Ships: Speech Recognition Accuracy Hits 91%

KEY FACTS

iFlytek Input has been upgraded to version 16.0.1, adding a third-generation voice input feature called Intent-to-Draft and a personalized AI keyboard.
The new version runs on the Spark-ASR-2.0 speech recognition large model, supports 202 dialects, and raises dialect recognition accuracy from 85% to 91%.
Recognition accuracy in high-noise environments rose from 66% to 84%, and accuracy on English slots in mixed Chinese-English speech rose from 33% to 80%.
The personalized AI keyboard has accumulated 300,000 expression skills contributed by more than 150,000 creators, covering over 10,000 niche scenarios.
The company says more than 60% of users have already taken part in shaping their own custom keyboard.

KEY DATA

91%Dialect recognition accuracy

84%High-noise environment recognition accuracy

80%Mixed Chinese-English English slot accuracy

300,000Number of expression skills

PHYSIX OBSERVATION

iFlytek is packing large-model memory and scenario skills into the keyboard, and the speech recognition metrics improve markedly, especially in high-noise and mixed Chinese-English scenarios. This means the input method is turning from a typing tool into an intent-understanding entry point, but the practicality of the skills ecosystem and the privacy boundary remain the key tests.

Source: ifanr report