qwenlm/qwen3-asr is an open-source speech recognition model family developed by the Qwen team at Alibaba Cloud. The repository contains Qwen3-ASR-1.7B and Qwen3-ASR-0.6B, along with Qwen3-ForcedAligner-0.6B. ASR stands for automatic speech recognition, which turns speech into text. According to the repository description, the two ASR models support language identification and speech recognition across 52 languages and dialects, and offer streaming and offline inference as well as long-form audio transcription.
Forced alignment is the task of fitting a given text to time positions in speech. Qwen3-ForcedAligner-0.6B performs this alignment in 11 languages. So this repository covers both the step of obtaining text from speech and the step of checking at which point in the audio that text occurs. The implementation language is Python, and the license is Apache-2.0.
However, the available records contain no recognition accuracy, processing speed, or comparison with other models. Music and singing recognition and timestamp prediction are also mentioned in the repository description, but the available README summary gives no operating conditions or performance evidence. The feature scope is therefore confirmed, but there is not enough basis to judge actual performance or the conditions under which it applies.