Alibaba just dropped Qwen-Audio-3.0-ASR-Flash, a speech recognition model with a knack for context. The company is betting it'll stand out in specialized fields, with improved recognition for medical and industrial terminology.
You can feed it custom hotwords — drug names, equipment models, industry acronyms — and the model will pay extra attention to them. It also cleans up speech, producing a transcript with punctuation and proper structure.
In Alibaba's internal tests, the model hit a 95.36% recall rate for medical terms and 93.24% for industrial terms. It's available through three interfaces: real-time streaming, recorded files, and batch (non-real-time) processing.
https://twitter.com/Alibaba_Qwen/status/2083111834123407825