说话人严重塌缩:参考 3 人,模型只输出 1 人;出现幻觉/插入内容:插入率达 34.8%,输出明显长于参考。DER@0.25 高达 314.8%,说话人归属与切分同时失效,导致对话结构难以恢复。
0 good grief kay okay 0 put it on in that way thanks oops i will wait until you are all hooked up
i will wait until you are all in the picket fence 0 get creative okay okay and she put in he has got a 0 okay put her in that way
说话人与时间戳对比
| 类型 | 说话人 | 开始 (s) | 结束 (s) | 文本 |
|---|---|---|---|---|
| GT | B | 0.000 | 2.806 | I'll wait until you're all um hooked up. |
| GT | D | 17.379 | 19.197 | Oh good grief. |
| GT | D | 22.689 | 23.295 | 'Kay. |
| GT | B | 29.611 | 30.520 | Okay. |
| GT | D | 49.827 | 50.685 | Oh. |
| GT | B | 52.113 | 54.976 | Put it on in that way. Thanks. |
| GT | A | 53.980 | 54.502 | Oops. |
| Pred | speaker1 | 0.000 | 2.120 | I'll wait until you're all in the picket fence. |
| Pred | speaker1 | 14.440 | 16.440 | Oh, get creative. |
| Pred | speaker1 | 18.440 | 18.840 | Okay. |
| Pred | speaker1 | 24.440 | 24.840 | Okay. |
| Pred | speaker1 | 44.440 | 46.440 | And she put in, he's got a... Oh. |
| Pred | speaker1 | 47.440 | 48.440 | Okay, put her in that way. |
查看原始模型输出 (prediction_raw)
{"segments":[{"index":1,"start":0.00,"end":2.12,"speaker":"speaker1","text":"I'll wait until you're all in the picket fence."},{"index":2,"start":14.44,"end":16.44,"speaker":"speaker1","text":"Oh, get creative."},{"index":3,"start":18.44,"end":18.84,"speaker":"speaker1","text":"Okay."},{"index":4,"start":24.44,"end":24.84,"speaker":"speaker1","text":"Okay."},{"index":5,"start":44.44,"end":46.44,"speaker":"speaker1","text":"And she put in, he's got a... Oh."},{"index":6,"start":47.44,"end":48.44,"speaker":"speaker1","text":"Okay, put her in that way."}]}/cpfs/user/jingpeng/workspace/nfs/data/test/sa_asr/test_sa_asr/audio/60s/ami_ES2004c_chunk_0000.wav