Alibaba’s Qwen team has introduced Qwen3-ASR-Flash, a new automatic speech recognition (ASR) service built on the intelligence of Qwen3-Omni and trained on tens of millions of hours of multimodal data. The model is designed to deliver accurate, flexible, and robust transcription across a wide range of languages, accents, and environments.

Key Capabilities

1. High Recognition Accuracy: Qwen3-ASR-Flash demonstrates strong performance across industry benchmarks, surpassing many competing models for Chinese, English, and nine other languages.
2. Support for 11 Languages: The model handles diverse languages and accents, including:
3. Chinese: Mandarin and major dialects such as Sichuanese, Minnan (Hokkien), Wu, and Cantonese.
4. English: British, American, and regional accents.
5. Others: French, German, Russian, Italian, Spanish, Portuguese, Japanese, Korean, and Arabic.

Contextual Biasing

Users can provide background text whether keyword lists, full documents, or mixed formats to guide the transcription toward domain-specific accuracy. This removes the need for manual preprocessing and enables tailored outputs.

Singing Voice Recognition

Qwen3-ASR-Flash can transcribe songs accurately, even in the presence of background music.

Noise Robustness

The system maintains performance under challenging acoustic conditions, rejecting non-speech sounds such as silence or environmental noise.

Continuous Development

As an API service, Qwen3-ASR-Flash will continue to evolve, with ongoing improvements to recognition accuracy and feature optimization. The Qwen team emphasizes updates that enhance both multilingual support and usability in real-world conditions.