AI speech transcription

Alibaba’s Qwen team introduces Qwen3-ASR-Flash, a new AI speech transcription model trained on millions of hours of speech. It boasts improved accuracy, especially in challenging acoustics and diverse languages, outperforming competitors in standard Chinese, regional accents, English, and music transcription. A key feature is flexible contextual biasing using varied input formats. Supporting 11 languages and regional dialects, Qwen3-ASR-Flash aims to be a global transcription tool, also filtering out non-speech segments.