Specification Parameters
Input
- Audio formats
- aac, amr, avi, flac, flv, m4a, mkv, mov, mp3, mp4, mpeg, ogg, opus, wav, webm, wma, wmv · Supported list for qwen-audio-3.1-asr-flash / filetrans
- File size and duration
- 同步 ≤2GB 且 ≤5 分钟;异步 ≤2GB 且 ≤12 小时 · Varies by model version
Interface
- Invocation method
- REST API,DashScope API Key 鉴权;同步支持 SSE 流式,异步提交任务后轮询 · Endpoint cn-beijing.maas.aliyuncs.com
Output
- Timestamps
- 句级与字级 begin_time/end_time,毫秒,始终开启 · Cannot be disabled
Features
- Speaker diarization
- speaker_diarization_enabled: true(同步);diarization_enabled: true(异步) · Requires mono audio
Region
- Available regions
- 中国(北京)、新加坡、美国(弗吉尼亚) · Qwen-Audio-ASR and Filetrans available in Beijing and Singapore
资料下载
Comparison with Similar Products
| Specifications | 本品 · qwen-audio-3.1-asr-flash | kimi-k3.1-pro | 火山方舟模型 API | doubao-seed-2.1-turbo | CreateArkOfficialResultQuery / GetArkOfficialResult |
|---|---|---|---|---|---|
| Main controller | — | — | — | — | — |
| NPU | — | — | — | — | — |
| Memory / VRAM | — | — | — | — | — |
| Reference unit price | Request a Quote | Request a Quote | Request a Quote | Request a Quote | ¥0.1 |
| Inventory | — | — | — | — | — |
Product introduction
Purpose and scenarios: Official documentation states this service converts recorded audio/video files into text for meeting transcription, call analysis, and subtitles. Short audio can return results in a single request, while long recordings are processed as async tasks.
Interface and invocation: Call REST endpoints using DASHSCOPE_API_KEY. The synchronous version qwen-audio-3.1-asr-flash accepts audio via URL or Base64 Data URI, and supports X-DashScope-SSE: enable streaming output; the filetrans version submits a task and polls status (PENDING/RUNNING/SUCCEEDED/FAILED/CANCELED), then downloads the transcription JSON upon success, with result URLs valid for 24 hours.
Input limits: The synchronous version supports formats including aac, amr, avi, flac, flv, m4a, mkv, mov, mp3, mp4, mpeg, ogg, opus, wav, webm, wma, wmv, with single files ≤2GB and ≤5 minutes; the async version is likewise ≤2GB but up to 12 hours. Speaker diarization requires mono audio, and when diarization is enabled, async transcription is recommended to stay under 2 hours to reduce failure or timeout risk.
Output capabilities: Results include full text, sentence-level and character-level timestamps (milliseconds), and speaker_id; timestamps are always enabled and cannot be turned off. Text polishing (cleaning up fillers, repetition, and self-corrections) is natively supported by qwen-audio-3.1-asr-flash; dialects are transcribed as Mandarin by default, and keep_dialect: true preserves dialect expressions. When speaker diarization is enabled, keep_dialect and language_hints do not take effect.
Pre-purchase checks: Confirm the required model is available in the target region (China Beijing / Singapore / US Virginia); confirm audio format and sample rate match the actual files; in production, handle audio storage, submission and query rate limits, completion callbacks, and result persistence yourself.
Compilation Notes 内容由 AI 根据公开资料整理;事实与推断分开呈现。
Content Correction