qwen3-asr-models — MAESTRO mirror

Verbatim mirror of two Alibaba Qwen releases (both Apache-2.0), laid out for MAESTRO's download manager:

Path	Upstream	What
1.7b/	Qwen/Qwen3-ASR-1.7B	Multilingual ASR (30 languages) built for speech, singing, and songs with backing music
aligner/	Qwen/Qwen3-ForcedAligner-0.6B	Companion forced aligner for word/character-level timestamps

Files are unmodified (safetensors, bf16); every LFS file's sha256 matches the upstream repos at mirror time. The 0.6B ASR variant is not mirrored — it lost MAESTRO's lyric-transcription bench to the 1.7B.

Used by MAESTRO's Transcribe surface and the SoulX-Singer preprocess lyric lane. All credit to the Qwen team — see the upstream model cards for the technical report and usage.