CoinWorld News, the Qianwen large model has officially upgraded to the real-time speech recognition model Fun-ASR-Realtime. The first-word latency of this model is controlled at the level of hundreds of milliseconds, recognition accuracy is close to offline models, and it supports 16 dialects and 30 languages.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 6
  • Repost
  • Share
Comment
Add a comment
Add a comment
GateUser-2eca626f
· 2h ago
The number of supported languages has doubled, right? Multilingual real-time transcription is crucial for overseas products.
View OriginalReply0
BluePeonyDarkroom
· 2h ago
Close to the accuracy of offline models, with latency compressed to under a hundred milliseconds. I'd like to dive deeper into the technical details.
View OriginalReply0
TheCandlestickChartLooksLikeAn
· 6h ago
16 dialects! Finally, I don't have to speak broken Mandarin to an AI.
View OriginalReply0
QuantsAndCats
· 6h ago
Qwen's latest upgrade is really something; it's not easy to balance real-time performance and accuracy.
View OriginalReply0
GateUser-94818fd0
· 6h ago
Coverage of 30 languages is quite broad. The name Fun-ASR-Realtime is quite interesting.
View OriginalReply0
Don'tLetTheContractScamMyMom.
· 6h ago
The latency of the first character by one hundred milliseconds is indeed great, and the support for 16 dialects for voice-interaction scenarios is extremely practical.
View OriginalReply0
  • Pinned