Let's build the GPT Tokenizer
说话人分离结果（匿名标签）。同目录的
transcript.md未作任何改动。 标签 → 真人姓名的映射属于编辑判断，请记在 wiki 视频页开头，见 CLAUDE.md。 分离在抢话/串场处会错配：标签是证据不是判决，与内容矛盾时以内容为准。
说话占比¶
| 说话人 | 时长 | 占比 |
|---|---|---|
| SPEAKER_00 | 02:12:23 | 100.0% |
| SPEAKER_01 | 00:00:02 | 0.0% |
（访谈里占比低的通常是主持人，可作为认人的线索之一。）
说话轮次¶
| 起 | 止 | 说话人 |
|---|---|---|
| 00:00:00 | 00:04:58 | SPEAKER_00 |
| 00:04:53 | 00:04:54 | SPEAKER_01 |
| 00:04:58 | 00:17:08 | SPEAKER_00 |
| 00:17:11 | 00:36:24 | SPEAKER_00 |
| 00:36:23 | 00:36:23 | SPEAKER_01 |
| 00:36:26 | 00:37:49 | SPEAKER_00 |
| 00:37:52 | 00:39:17 | SPEAKER_00 |
| 00:39:20 | 00:44:33 | SPEAKER_00 |
| 00:44:36 | 00:46:56 | SPEAKER_00 |
| 00:46:52 | 00:46:52 | SPEAKER_01 |
| 00:46:56 | 00:47:41 | SPEAKER_00 |
| 00:47:44 | 00:49:53 | SPEAKER_00 |
| 00:49:55 | 00:51:10 | SPEAKER_00 |
| 00:51:12 | 00:53:56 | SPEAKER_00 |
| 00:53:59 | 00:54:18 | SPEAKER_00 |
| 00:54:20 | 00:54:31 | SPEAKER_00 |
| 00:54:27 | 00:54:27 | SPEAKER_01 |
| 00:54:33 | 00:56:02 | SPEAKER_00 |
| 00:56:05 | 01:04:32 | SPEAKER_00 |
| 01:04:30 | 01:04:31 | SPEAKER_01 |
| 01:04:32 | 01:05:24 | SPEAKER_00 |
| 01:05:26 | 01:18:57 | SPEAKER_00 |
| 01:18:59 | 01:19:39 | SPEAKER_00 |
| 01:19:41 | 01:27:59 | SPEAKER_00 |
| 01:27:56 | 01:27:56 | SPEAKER_01 |
| 01:28:00 | 01:32:07 | SPEAKER_00 |
| 01:32:10 | 01:32:59 | SPEAKER_00 |
| 01:33:01 | 01:36:29 | SPEAKER_00 |
| 01:36:32 | 01:37:23 | SPEAKER_00 |
| 01:37:25 | 01:39:25 | SPEAKER_00 |
| 01:39:18 | 01:39:19 | SPEAKER_01 |
| 01:39:26 | 01:40:56 | SPEAKER_00 |
| 01:40:55 | 01:40:55 | SPEAKER_01 |
| 01:40:57 | 01:43:22 | SPEAKER_00 |
| 01:43:20 | 01:43:20 | SPEAKER_01 |
| 01:43:22 | 01:50:18 | SPEAKER_00 |
| 01:50:13 | 01:50:13 | SPEAKER_01 |
| 01:50:18 | 01:59:07 | SPEAKER_00 |
| 01:59:04 | 01:59:04 | SPEAKER_01 |
| 01:59:08 | 02:06:22 | SPEAKER_00 |
| 02:06:19 | 02:06:20 | SPEAKER_01 |
| 02:06:22 | 02:09:37 | SPEAKER_00 |
| 02:09:21 | 02:09:21 | SPEAKER_01 |
| 02:09:38 | 02:12:47 | SPEAKER_00 |
| 02:12:54 | 02:13:11 | SPEAKER_00 |
| 02:13:16 | 02:13:28 | SPEAKER_00 |