Today Qianwen upgraded its real-time speech-recognition model Fun-ASR-Realtime,
a streaming ASR system with first-word latency controlled at the
hundred-millisecond level and recognition accuracy approaching offline models;
it supports 16 dialects and 30 languages.