Qwen ASR · Introduction
Qwen speech recognition (ASR) uses the qwen gateway; pick the concrete model via the request body model field. Transcribes audio to text, billed by input audio seconds (0.022 credits/sec); output is free. Callers only pass audioUrl + duration (audio seconds); all 4 models return a taskId on submit and are polled by taskId — they differ only in result form: • File transcription: fun-asr / fun-asr-mtl (multilingual) / qwen3-asr-flash-filetrans — on succeeded, resultUrl is the transcription file URL (JSON, valid 24h, with text and word-level timestamps). • Fast recognition: qwen3-asr-flash — on succeeded, resultUrl is the recognized text directly (no timestamps).
How to call
- Submit:
POST /api/cqt/generator/qwen, returns a taskId. - Query:
GET /api/cqt/info/qwen?id=taskId, poll until status = succeeded.
Capability map
Common Errors
| Code | msg | Meaning | How to fix |
|---|---|---|---|
| 400 | audioUrl required / Unsupported model | audioUrl is empty, or model is not one of the 4 ASR models | Pass a valid audio URL; use model fun-asr / fun-asr-mtl / qwen3-asr-flash / qwen3-asr-flash-filetrans |
| 500 | failed / timeout | Upstream recognition failed or timed out; credits auto-refunded, see data.errorMsg | Check audio accessibility/format and retry |
Errors are returned as upstream bizCode, e.g. upstream bizCode=451, msg=prompt_unsafe.
Tip: after saving your API Key via Authorize, every capability page supports "Try it".