# 创建语音妙记任务

提交可公网访问的录音/录像 URL，立即返回任务 ID。用 `GET /v1/audio/minutes/{taskId}` 轮询（建议间隔 ≥30 秒）。

**仅中国站。** 两条模型共用同一接口，换 `model` 即可。

- `xrtoken-minutes`（会议纪要）：录音文件识别 2.0 + DeepSeek V4 Flash 总结。只需 `model` + `file_url`。返回完整转写与结构化 `report`（version=2），含核心速览、分层总结、章节、待办、决策、金句、待核对项及按内容选择的图形数据。兼容保留 `summary` / `chapters` / `extraction`。`features` / `all_activate` 不生效。识别按时长计费，总结按 Flash token，价格见模型市场。
- `doubao-minutes`（豆包语音妙记）：火山官方妙记。转写必开，且至少再开一项附加能力，或 `all_activate=true` 打包全部（总结/章节/待办/问答/翻译）。按时长计费，价格见模型市场。

`file_url` 须 http(s)，上游要能拉到（可用 `POST /v1/files` 上传后用返回的 URL）。文件 < 1GB、时长 ≤ 2 小时。提交时预冻，按已完成识别时长及已观测总结 token 结算，释放余额。`xrtoken-minutes` 识别成功后，总结独立最多等待 30 分钟；摘要失败保留转写并按已发生用量结算，不自动重试。识别失败释放预冻。

## POST /v1/audio/minutes

> Create a meeting-minutes task

Submit a publicly reachable audio/video URL and receive a task ID immediately. Poll `GET /v1/audio/minutes/{taskId}` (interval ≥ 30s).

**CN only.** Same endpoints; switch `model`.

- `xrtoken-minutes`: file ASR 2.0 + DeepSeek V4 Flash summary. Only `model` + `file_url` are required. Returns full transcription and report (version=2): overview, key points, nested topics, chapters, todos, decisions, verbatim quotes, review items and content-appropriate visual data. Legacy summary/chapters/extraction remain available. `features` / `all_activate` are ignored. ASR is billed by duration; summary by Flash tokens. See the model marketplace for rates.
- `doubao-minutes`: official Lark Minutes. Transcription plus at least one extra, or `all_activate=true` for summary/chapters/todos/qa/translate. Billed by duration. See the model marketplace for rates.

`file_url` must be http(s) and fetchable by the upstream (you can `POST /v1/files` first). File < 1GB, duration ≤ 2 hours. Reserve on submit; settle completed ASR and observed summary tokens, releasing the remainder. For `xrtoken-minutes`, summary generation has its own 30-minute limit after ASR succeeds. A failed summary retains transcription and settles observed usage without automatic retry. ASR failure releases the reservation.

### Authentication

`Authorization: Bearer tr-xxx`

### Request Body

Content-Type: `application/json`

- **model** `string` **(required)**  
  Minutes model ID. `doubao-minutes` (official) or `xrtoken-minutes` (ASR 2.0 + Flash summary)
- **file_url** `string` **(required)**  
  Publicly reachable audio or video URL (< 1GB, ≤ 2 hours)
- **recorded_at** `string,null`  
  Optional recording start in RFC3339 with offset. Stored in xrtoken-minutes report.metadata; never substituted with task creation time.
- **timezone** `string,null`  
  Optional IANA timezone such as Asia/Shanghai; null when absent.
- **file_type** ``audio` | `video``  
  Inferred from the URL suffix when omitted
- **source_lang** ``zh_cn` | `en_us`` (default: `zh_cn`)  
  
- **all_activate** `boolean` (default: `false`)  
  doubao-minutes only. Enables every extra. Ignored for xrtoken-minutes.
- **speaker_identification** `boolean` (default: `true`)  
  Speaker diarization. Always on for xrtoken-minutes.
- **number_of_speakers** `integer` (default: `0`)  
  0 lets the model detect speaker count
- **hot_words** `string`  
  
- **need_word_time_series** `boolean` (default: `false`)  
  
- **features** `object`  
  Extra features for `doubao-minutes` only. At least one is required unless `all_activate` is true. Ignored for `xrtoken-minutes` (the complete structured report is always on).
- **features.summary** `boolean`  
  
- **features.chapters** `boolean`  
  
- **features.todos** `boolean`  
  
- **features.qa** `boolean`  
  
- **features.translate** `boolean`  
  
- **features.target_lang** ``zh_cn` | `en_us``  
  

### Response

- **id** `string` **(required)**  
  
- **request_id** `string`  
  
- **model** `string` **(required)**  
  
- **status** ``processing`` **(required)**  
  
- **created_at** `string` **(required)**  
  

### Error Codes

- `400`: 
- `401`: 
- `402`: 
- `429`: 
- `502`: