1-ASR-Flash-Next, retaining its audio encoder, audio-to-language adaptor and mixture-of-experts language decoder.
Direct decision head makes bounded audio judgments about 20 times faster
Qwen-Audio-Jev replaces generated answers with typed outputs for categorical choices, binary judgments and numerical ratings.
Chinese Tech
Ying Fang · Zhifu Gao · Qing Gu · Jiajun He · Kangqi Jing · Xiangang Li · +2 more
Token Foundry · Alibaba Group
Research Digest··2 min read
Fang et al.
§