apizt · AUDIO LAB / REVIEW
Whisper 长音频分块转录
实际 ASR 输出 · 原始误识别保留03 / 逐项核对与输出证据
| 文件 / 项目 | 时间 / 状态 | 原始结果 / 核对依据 |
|---|---|---|
| chunk-000.json | 0.00–5.40 | First, copy the recording into a new folder, keep the original file unchanged. |
| chunk-000.json | 7.02–12.62 | Next, transcribe the second section, if the task fails, save the error message. |
| chunk-000.json | 14.04–14.98 | Finally, come back. |
| chunk-001.json | 0.00–3.72 | find the sections in order, check the boundary before publishing. |
| chunk-001.json | 5.28–10.38 | First, copy the recording into a new folder, keep the original file unchanged. |
| chunk-001.json | 11.68–14.98 | Next, transcribe the second section, if the... |
| chunk-002.json | 0.00–2.62 | task fails, save the error message. |
| chunk-002.json | 4.12–8.58 | Finally, combine the sections in order, check the boundary before publishing. |
| combined.json | 0.00–5.40 | First, copy the recording into a new folder, keep the original file unchanged. |
真实边界错例：切断 combine 后出现 come back / find。结构通过不代表语义完整，应回听并重切对应边界。
04 / 本页验收
□ 切片 offset 和源 hash 可追溯且覆盖完整。
□ 合并稿按原音绝对时间排序。
□ 每个切片边界已回听，结构通过不掩盖半词。