运行报错！求助

[LittleMaWen]

您好，我正在复现您的实验，unbuntu16.04，python3.7，我首先运行了train.py文件，出现了以下错误，请问该如何解决？诚盼回复
model_build_time 5.370615005493164
get batch time 1.98e-05s
forward process time 7.57s
beginning to select..........
select best batch time 0.188s
select_batch_time: 7.82932448387146
Traceback (most recent call last):
File "train.py", line 181, in
main()
File "train.py", line 125, in main
loss = model.train_on_batch(x, y)
File "/home/dcase/miniconda3/lib/python3.7/site-packages/keras/engine/training.py", line 1808, in train_on_batch
check_batch_axis=True)
File "/home/dcase/miniconda3/lib/python3.7/site-packages/keras/engine/training.py", line 1411, in _standardize_user_data
exception_prefix='target')
File "/home/dcase/miniconda3/lib/python3.7/site-packages/keras/engine/training.py", line 153, in _standardize_input_data
str(array.shape))
ValueError: Error when checking target: expected ln to have shape (None, 512) but got array with shape (96, 1)

[Walleclipse]

根据issue 7, 需要把 train.py 中的 121行改成以下形式:
y = np.random.uniform(size=(x.shape[0], 512))

[LittleMaWen]

谢谢谢谢您！解决了！现在正在运行！感谢！另外我在刚刚关闭的问题里问过您了，就是首先python train.py之后，接下来的步骤是怎样？train.py是在训练模型嘛？我在实验报告里也没有找到详细的运行步骤，烦请回复，谢谢您！

[Walleclipse]

train.py 就是训练模型。之后你可以用 test_model.py进行模型的测试。
本项目训练出的模型是一个speaker embedding 模型，也就是输入一段语音，输出一个512维的embedding表示，此表示可以用于说话者验证或者识别等不同领域。
接下来我不太清楚你要达到的目的是什么？
有关训练流程和画出误差曲线你可以查看 issue 1
有关训练后如何进行 inference 你可以查看 issue 30
有关声纹识别你可以查看 issue 21

[LittleMaWen]

嗷嗷~谢谢您！我昨天早晨修正错误后就开始运行了，今早发现运行过程中出现了错误，具体如下：
OSError: [Errno 30] Read-only file system

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "train.py", line 181, in
File "train.py", line 134, in main
OSError: [Errno 30] Read-only file system
(tensorflow) dcase@dcase-PowerEdge-R730:~/mawen/SW/Deep_Speaker-speaker_recognition_system-master$
Broadcast message from systemd-journald@dcase-PowerEdge-R730 (Wed 2019-08-14 07:37:30 CST):

systemd[1]: Caught , dumped core as pid 18055.

Broadcast message from systemd-journald@dcase-PowerEdge-R730 (Wed 2019-08-14 07:37:30 CST):

systemd[1]: Freezing execution.
请问您知道是什么原因么？

[Walleclipse]

我不太清楚。你报错的代码块是写入log的那一段，可能是你的操作系统没有写的权限？
你可以查看 pypa/virtualenv#209 或者 https://askubuntu.com/questions/1082876/oserror-errno-30-read-only-file-system 或者 https://forums.fast.ai/t/how-can-i-load-a-pretrained-model-on-kaggle-using-fastai/13941/4

[LittleMaWen]

OKOK谢谢您！非常感谢您不厌其烦为我解答谢谢您！

[LittleMaWen]

嗯嗯~学长，这个问题已经解决了，是我自己系统的问题。
现在我已经python train.py两天多了，程序还在运行，
2019-08-19 09:07:35,157 [INFO] train.py/main | == Presenting step #132313
2019-08-19 09:07:35,351 [INFO] train.py/main | == Processed in 0.19s by the network, training loss = 0.004015110433101654.
get batch time 4.53e-06s
forward process time 0.317s
beginning to select..........
select best batch time 0.71s
select_batch_time: 1.0472040176391602
这是最近一条显示结果，我看到您在其他问题中回答train.py是个无限循环的，需要自己kill掉，请问我现在这个结果是可以kill掉了吗？我不是很明白

[LittleMaWen]

学长，我在训练到下面这个结果时kill掉了这个程序，
2019-08-20 12:24:52,587 [INFO] train.py/main | == Presenting step #211143
2019-08-20 12:24:52,776 [INFO] train.py/main | == Processed in 0.19s by the network, training loss = 0.0009784128051251173.
然后按照学长说的test_model.py了一下，得到了这个结果，
Found checkpoint [checkpoints/model_211000_0.00100.h5]. Resume from here...
Found 0000368 files with 00003 different speakers.
f-measure = 0.9999999999995, true positive rate = 1.0, accuracy = 0.99999999999998, equal error rate = 0.0
学长，请问是这样吗？

[Walleclipse]

应该是对的。但你这个是在完整的LibriSpeech数据上跑的还是我这个repo里面的LibriSpeechSamples上跑的？
你需要用完整的数据集。

[LittleMaWen]

是的学长，谢谢您回复，我现在才打算用完整的数据集重新跑一遍。我下载了train-clean-100和train-clean-360，我看实验报告，如果我没有理解错的话，请问是否应该先pre_process.py 得到train-clean-100-npy，然后再train.py?
另外就是实验报告中提到的test-clean数据集，请问这是作为测试集的吗？它在哪一部分使用？在跑LibriSpeechSamples时，我也只看到用了train-clean-100中的一部分音频，没有找到test-clean的部分。

[Walleclipse]

1. 是先通过 pre_process.py 来预处理数据，随后再train
2. test_clean 数据集也是从 Librispeech 这里下载的，test_clean是专门用来测试的数据

[LittleMaWen]

学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果：
17200,1.8035286664962769
17201,0.8399255871772766
17202,0.7849252820014954
17203,0.20637908577919006
......
200847,0.0009674201137386262
200848,0.0009672052692621946
200849,0.0009669908322393894
......
382718,0.2163151204586029
382719,0.2672955095767975
382720,0.5557892322540283
得到的train_acc_eer.txt结果：
17200,0.0,0.9999999999995,0.99999999999998
17210,0.0,0.9999999999995,0.99999999999998
......
211130,0.0,0.9999999999995,0.99999999999998
211140,0.0,0.9999999999995,0.99999999999998
211000,0.32981049562682213,0.1791044776114538,0.9607142857142851
211000,0.282798833819242,0.1739130434777845,0.9728571428571422
......
382440,0.03024781341107871,0.9152542372876369,0.9964285714285708
382450,0.008746355685131196,0.9285714285709286,0.9971428571428564
得到的acc_eer.txt结果：
17200,0.0,0.9999999999995,0.99999999999998
17200,0.0,0.9999999999995,0.99999999999998
......
210800,0.0,0.9999999999995,0.99999999999998
211000,0.0,0.9999999999995,0.99999999999998
211000,0.3589743589743589,0.12269938650270346,0.9266666666666662
211000,0.358451072736787,0.10714285714243434,0.9743589743589739
......
382200,0.0567765567765568,0.7297297297292311,0.9897435897435893
382400,0.13291470434327576,0.6666666666661832,0.9887179487179483
这时候我就kill掉了train.py，然后python test_model.py得到了结果：
f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261
请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？

[MAGUADIDI]

你好，我也想复现这个实验，可以指导一下吗？

[izhangy]

你好，在pre_training.py运行时出现：Found 0000368 files with 00003 different speakers.，一直不动了，请问是怎么回事？

[MAGUADIDI]

没有遇到这个问题。从字面理解，你的人数太少了，而且样本也有些少。

…

---原始邮件--- 发件人: "izhangy"<[email protected]&gt; 发送时间: 2019年12月3日(星期二) 中午11:07 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送: "Comment"<[email protected]&gt;;"QI ZHANG"<[email protected]&gt;; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 你好，在pre_training.py运行时出现：Found 0000368 files with 00003 different speakers.，一直不动了，请问是怎么回事？ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[izhangy]

非常感谢您的回答！我增加点数据量再试试
还有一个问题，运行train.py后没有“checkpoints‘文件生成，请问这也是跟数据量有关吗？

[MAGUADIDI]

这个我不清楚，应该是无关的。只要训练了就有模型生成 ------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"izhangy"<[email protected]&gt;; 发送时间:&nbsp;2019年12月3日(星期二) 下午2:44 收件人:&nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送:&nbsp;"Popcorn"<[email protected]&gt;;"Comment"<[email protected]&gt;; 主题:&nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 非常感谢您的回答！我增加点数据量再试试 还有一个问题，运行train.py后没有“checkpoints‘文件生成，请问这也是跟数据量有关吗？ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[izhangy]

“程序运行两天了还没结束，是怎么回事呢？"
您好，跟楼主一样，我也遇到了这样的问题，请问怎么解决？

[MAGUADIDI]

你就光运行不看代码吗？程序本来就是死循环，我是得到合适的模型之后自己手动结束。

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"izhangy"<[email protected]&gt;; 发送时间:&nbsp;2019年12月6日(星期五) 上午10:13 收件人:&nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送:&nbsp;"Popcorn"<[email protected]&gt;;"Comment"<[email protected]&gt;; 主题:&nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) “程序运行两天了还没结束，是怎么回事呢？" 您好，跟楼主一样，我也遇到了这样的问题，请问怎么解决？ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[izhangy]

见笑了，我调通后就跑了，也没看------------------ 原始邮件 ------------------ 发件人: "QI ZHANG"<[email protected]> 发送时间: 2019年12月6日(星期五) 上午10:19 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]>; 抄送: "izhangy"<[email protected]>;"Comment"<[email protected]>; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 你就光运行不看代码吗？程序本来就是死循环，我是得到合适的模型之后自己手动结束。

…

------------------&nbsp;原始邮&#x4EF6;&nbsp;------------------ 发&#x4EF6;人:&nbsp;"izhangy"<[email protected]&gt;; 发送时间:&nbsp;2019年12月6日(星期五) 上午10:13 收&#x4EF6;人:&nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送:&nbsp;"Popcorn"<[email protected]&gt;;"Comment"<[email protected]&gt;; 主题:&nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) “程序运行两天了还没结束，是怎么回事呢？" 您好，跟楼主一样，我也遇到了这样的问题，请问怎么解决？ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. &mdash; You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[yy835055664]

你好学长，我看了你复现那个Deep-speaker，也出结果了，我想问你几个问题？
1.checkpiont里面.h5文件怎么产生的？
2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？
望学长能给予解答，谢谢！

[yy835055664]

学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果：
17200,1.8035286664962769
17201,0.8399255871772766
17202,0.7849252820014954
17203,0.20637908577919006
......
200847,0.0009674201137386262
200848,0.0009672052692621946
200849,0.0009669908322393894
......
382718,0.2163151204586029
382719,0.2672955095767975
382720,0.5557892322540283
得到的train_acc_eer.txt结果：
17200,0.0,0.9999999999995,0.99999999999998
17210,0.0,0.9999999999995,0.99999999999998
......
211130,0.0,0.9999999999995,0.99999999999998
211140,0.0,0.9999999999995,0.99999999999998
211000,0.32981049562682213,0.1791044776114538,0.9607142857142851
211000,0.282798833819242,0.1739130434777845,0.9728571428571422
......
382440,0.03024781341107871,0.9152542372876369,0.9964285714285708
382450,0.008746355685131196,0.9285714285709286,0.9971428571428564
得到的acc_eer.txt结果：
17200,0.0,0.9999999999995,0.99999999999998
17200,0.0,0.9999999999995,0.99999999999998
......
210800,0.0,0.9999999999995,0.99999999999998
211000,0.0,0.9999999999995,0.99999999999998
211000,0.3589743589743589,0.12269938650270346,0.9266666666666662
211000,0.358451072736787,0.10714285714243434,0.9743589743589739
......
382200,0.0567765567765568,0.7297297297292311,0.9897435897435893
382400,0.13291470434327576,0.6666666666661832,0.9887179487179483
这时候我就kill掉了train.py，然后python test_model.py得到了结果：
f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261
请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？

学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果：
17200,1.8035286664962769
17201,0.8399255871772766
17202,0.7849252820014954
17203,0.20637908577919006
......
200847,0.0009674201137386262
200848,0.0009672052692621946
200849,0.0009669908322393894
......
382718,0.2163151204586029
382719,0.2672955095767975
382720,0.5557892322540283
得到的train_acc_eer.txt结果：
17200,0.0,0.9999999999995,0.99999999999998
17210,0.0,0.9999999999995,0.99999999999998
......
211130,0.0,0.9999999999995,0.99999999999998
211140,0.0,0.9999999999995,0.99999999999998
211000,0.32981049562682213,0.1791044776114538,0.9607142857142851
211000,0.282798833819242,0.1739130434777845,0.9728571428571422
......
382440,0.03024781341107871,0.9152542372876369,0.9964285714285708
382450,0.008746355685131196,0.9285714285709286,0.9971428571428564
得到的acc_eer.txt结果：
17200,0.0,0.9999999999995,0.99999999999998
17200,0.0,0.9999999999995,0.99999999999998
......
210800,0.0,0.9999999999995,0.99999999999998
211000,0.0,0.9999999999995,0.99999999999998
211000,0.3589743589743589,0.12269938650270346,0.9266666666666662
211000,0.358451072736787,0.10714285714243434,0.9743589743589739
......
382200,0.0567765567765568,0.7297297297292311,0.9897435897435893
382400,0.13291470434327576,0.6666666666661832,0.9887179487179483
这时候我就kill掉了train.py，然后python test_model.py得到了结果：
f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261
请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？

你好学长，我看了你复现那个Deep-speaker，也出结果了，我想问你几个问题？
1.checkpiont里面.h5文件怎么产生的？
2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？
望学长能给予解答，谢谢！

[MAGUADIDI]

util只有函数，你要调用它。训练的久不久，还是要看数据，数据你觉得差不多就行

…

---原始邮件--- 发件人: "yy835055664"<[email protected]&gt; 发送时间: 2019年12月10日(星期二) 下午4:52 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送: "Comment"<[email protected]&gt;;"QI ZHANG"<[email protected]&gt;; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果： 17200,1.8035286664962769 17201,0.8399255871772766 17202,0.7849252820014954 17203,0.20637908577919006 ...... 200847,0.0009674201137386262 200848,0.0009672052692621946 200849,0.0009669908322393894 ...... 382718,0.2163151204586029 382719,0.2672955095767975 382720,0.5557892322540283 得到的train_acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17210,0.0,0.9999999999995,0.99999999999998 ...... 211130,0.0,0.9999999999995,0.99999999999998 211140,0.0,0.9999999999995,0.99999999999998 211000,0.32981049562682213,0.1791044776114538,0.9607142857142851 211000,0.282798833819242,0.1739130434777845,0.9728571428571422 ...... 382440,0.03024781341107871,0.9152542372876369,0.9964285714285708 382450,0.008746355685131196,0.9285714285709286,0.9971428571428564 得到的acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17200,0.0,0.9999999999995,0.99999999999998 ...... 210800,0.0,0.9999999999995,0.99999999999998 211000,0.0,0.9999999999995,0.99999999999998 211000,0.3589743589743589,0.12269938650270346,0.9266666666666662 211000,0.358451072736787,0.10714285714243434,0.9743589743589739 ...... 382200,0.0567765567765568,0.7297297297292311,0.9897435897435893 382400,0.13291470434327576,0.6666666666661832,0.9887179487179483 这时候我就kill掉了train.py，然后python test_model.py得到了结果： f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261 请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？ 学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果： 17200,1.8035286664962769 17201,0.8399255871772766 17202,0.7849252820014954 17203,0.20637908577919006 ...... 200847,0.0009674201137386262 200848,0.0009672052692621946 200849,0.0009669908322393894 ...... 382718,0.2163151204586029 382719,0.2672955095767975 382720,0.5557892322540283 得到的train_acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17210,0.0,0.9999999999995,0.99999999999998 ...... 211130,0.0,0.9999999999995,0.99999999999998 211140,0.0,0.9999999999995,0.99999999999998 211000,0.32981049562682213,0.1791044776114538,0.9607142857142851 211000,0.282798833819242,0.1739130434777845,0.9728571428571422 ...... 382440,0.03024781341107871,0.9152542372876369,0.9964285714285708 382450,0.008746355685131196,0.9285714285709286,0.9971428571428564 得到的acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17200,0.0,0.9999999999995,0.99999999999998 ...... 210800,0.0,0.9999999999995,0.99999999999998 211000,0.0,0.9999999999995,0.99999999999998 211000,0.3589743589743589,0.12269938650270346,0.9266666666666662 211000,0.358451072736787,0.10714285714243434,0.9743589743589739 ...... 382200,0.0567765567765568,0.7297297297292311,0.9897435897435893 382400,0.13291470434327576,0.6666666666661832,0.9887179487179483 这时候我就kill掉了train.py，然后python test_model.py得到了结果： f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261 请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？ 你好学长，我看了你复现那个Deep-speaker，也出结果了，我想问你几个问题？ 1.checkpiont里面.h5文件怎么产生的？ 2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？ 望学长能给予解答，谢谢！ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[yy835055664]

你好学长： 我想请教你几个问题？ 1.checkpiont里面.h5文件怎么产生的？ 2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？ 望学长能给予解答，谢谢！

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"QI ZHANG"<[email protected]&gt;; 发送时间:&nbsp;2019年12月10日(星期二) 下午5:04 收件人:&nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送:&nbsp;"未、、敢伤"<[email protected]&gt;;"Comment"<[email protected]&gt;; 主题:&nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) util只有函数，你要调用它。训练的久不久，还是要看数据，数据你觉得差不多就行

---原始邮件--- 发件人: "yy835055664"<[email protected]&amp;gt; 发送时间: 2019年12月10日(星期二) 下午4:52 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&amp;gt;; 抄送: "Comment"<[email protected]&amp;gt;;"QI ZHANG"<[email protected]&amp;gt;; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果： 17200,1.8035286664962769 17201,0.8399255871772766 17202,0.7849252820014954 17203,0.20637908577919006 ...... 200847,0.0009674201137386262 200848,0.0009672052692621946 200849,0.0009669908322393894 ...... 382718,0.2163151204586029 382719,0.2672955095767975 382720,0.5557892322540283 得到的train_acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17210,0.0,0.9999999999995,0.99999999999998 ...... 211130,0.0,0.9999999999995,0.99999999999998 211140,0.0,0.9999999999995,0.99999999999998 211000,0.32981049562682213,0.1791044776114538,0.9607142857142851 211000,0.282798833819242,0.1739130434777845,0.9728571428571422 ...... 382440,0.03024781341107871,0.9152542372876369,0.9964285714285708 382450,0.008746355685131196,0.9285714285709286,0.9971428571428564 得到的acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17200,0.0,0.9999999999995,0.99999999999998 ...... 210800,0.0,0.9999999999995,0.99999999999998 211000,0.0,0.9999999999995,0.99999999999998 211000,0.3589743589743589,0.12269938650270346,0.9266666666666662 211000,0.358451072736787,0.10714285714243434,0.9743589743589739 ...... 382200,0.0567765567765568,0.7297297297292311,0.9897435897435893 382400,0.13291470434327576,0.6666666666661832,0.9887179487179483 这时候我就kill掉了train.py，然后python test_model.py得到了结果： f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261 请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？ 学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果： 17200,1.8035286664962769 17201,0.8399255871772766 17202,0.7849252820014954 17203,0.20637908577919006 ...... 200847,0.0009674201137386262 200848,0.0009672052692621946 200849,0.0009669908322393894 ...... 382718,0.2163151204586029 382719,0.2672955095767975 382720,0.5557892322540283 得到的train_acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17210,0.0,0.9999999999995,0.99999999999998 ...... 211130,0.0,0.9999999999995,0.99999999999998 211140,0.0,0.9999999999995,0.99999999999998 211000,0.32981049562682213,0.1791044776114538,0.9607142857142851 211000,0.282798833819242,0.1739130434777845,0.9728571428571422 ...... 382440,0.03024781341107871,0.9152542372876369,0.9964285714285708 382450,0.008746355685131196,0.9285714285709286,0.9971428571428564 得到的acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17200,0.0,0.9999999999995,0.99999999999998 ...... 210800,0.0,0.9999999999995,0.99999999999998 211000,0.0,0.9999999999995,0.99999999999998 211000,0.3589743589743589,0.12269938650270346,0.9266666666666662 211000,0.358451072736787,0.10714285714243434,0.9743589743589739 ...... 382200,0.0567765567765568,0.7297297297292311,0.9897435897435893 382400,0.13291470434327576,0.6666666666661832,0.9887179487179483 这时候我就kill掉了train.py，然后python test_model.py得到了结果： f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261 请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？ 你好学长，我看了你复现那个Deep-speaker，也出结果了，我想问你几个问题？ 1.checkpiont里面.h5文件怎么产生的？ 2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？ 望学长能给予解答，谢谢！ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[MAGUADIDI]

训练就会产生。具体路径和你的代码有关系。代码有一个路径。我没有复现这个，只是看了看。

…

---原始邮件--- 发件人: "yy835055664"<[email protected]&gt; 发送时间: 2019年12月10日(星期二) 下午5:08 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送: "Comment"<[email protected]&gt;;"QI ZHANG"<[email protected]&gt;; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 你好学长： 我想请教你几个问题？ 1.checkpiont里面.h5文件怎么产生的？ 2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？ 望学长能给予解答，谢谢！

------------------&amp;nbsp;原始邮件&amp;nbsp;------------------ 发件人:&amp;nbsp;"QI ZHANG"<[email protected]&amp;gt;; 发送时间:&amp;nbsp;2019年12月10日(星期二) 下午5:04 收件人:&amp;nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&amp;gt;; 抄送:&amp;nbsp;"未、、敢伤"<[email protected]&amp;gt;;"Comment"<[email protected]&amp;gt;; 主题:&amp;nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) util只有函数，你要调用它。训练的久不久，还是要看数据，数据你觉得差不多就行

---原始邮件--- 发件人: "yy835055664"<[email protected]&amp;amp;gt; 发送时间: 2019年12月10日(星期二) 下午4:52 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&amp;amp;gt;; 抄送: "Comment"<[email protected]&amp;amp;gt;;"QI ZHANG"<[email protected]&amp;amp;gt;; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果： 17200,1.8035286664962769 17201,0.8399255871772766 17202,0.7849252820014954 17203,0.20637908577919006 ...... 200847,0.0009674201137386262 200848,0.0009672052692621946 200849,0.0009669908322393894 ...... 382718,0.2163151204586029 382719,0.2672955095767975 382720,0.5557892322540283 得到的train_acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17210,0.0,0.9999999999995,0.99999999999998 ...... 211130,0.0,0.9999999999995,0.99999999999998 211140,0.0,0.9999999999995,0.99999999999998 211000,0.32981049562682213,0.1791044776114538,0.9607142857142851 211000,0.282798833819242,0.1739130434777845,0.9728571428571422 ...... 382440,0.03024781341107871,0.9152542372876369,0.9964285714285708 382450,0.008746355685131196,0.9285714285709286,0.9971428571428564 得到的acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17200,0.0,0.9999999999995,0.99999999999998 ...... 210800,0.0,0.9999999999995,0.99999999999998 211000,0.0,0.9999999999995,0.99999999999998 211000,0.3589743589743589,0.12269938650270346,0.9266666666666662 211000,0.358451072736787,0.10714285714243434,0.9743589743589739 ...... 382200,0.0567765567765568,0.7297297297292311,0.9897435897435893 382400,0.13291470434327576,0.6666666666661832,0.9887179487179483 这时候我就kill掉了train.py，然后python test_model.py得到了结果： f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261 请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？ 学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果： 17200,1.8035286664962769 17201,0.8399255871772766 17202,0.7849252820014954 17203,0.20637908577919006 ...... 200847,0.0009674201137386262 200848,0.0009672052692621946 200849,0.0009669908322393894 ...... 382718,0.2163151204586029 382719,0.2672955095767975 382720,0.5557892322540283 得到的train_acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17210,0.0,0.9999999999995,0.99999999999998 ...... 211130,0.0,0.9999999999995,0.99999999999998 211140,0.0,0.9999999999995,0.99999999999998 211000,0.32981049562682213,0.1791044776114538,0.9607142857142851 211000,0.282798833819242,0.1739130434777845,0.9728571428571422 ...... 382440,0.03024781341107871,0.9152542372876369,0.9964285714285708 382450,0.008746355685131196,0.9285714285709286,0.9971428571428564 得到的acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17200,0.0,0.9999999999995,0.99999999999998 ...... 210800,0.0,0.9999999999995,0.99999999999998 211000,0.0,0.9999999999995,0.99999999999998 211000,0.3589743589743589,0.12269938650270346,0.9266666666666662 211000,0.358451072736787,0.10714285714243434,0.9743589743589739 ...... 382200,0.0567765567765568,0.7297297297292311,0.9897435897435893 382400,0.13291470434327576,0.6666666666661832,0.9887179487179483 这时候我就kill掉了train.py，然后python test_model.py得到了结果： f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261 请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？ 你好学长，我看了你复现那个Deep-speaker，也出结果了，我想问你几个问题？ 1.checkpiont里面.h5文件怎么产生的？ 2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？ 望学长能给予解答，谢谢！ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[yy835055664]

学长你好： 想问一下你知道这个问题这么解决吗？训练阶段

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"QI ZHANG"<[email protected]&gt;; 发送时间:&nbsp;2019年12月10日(星期二) 下午5:11 收件人:&nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送:&nbsp;"未、、敢伤"<[email protected]&gt;;"Comment"<[email protected]&gt;; 主题:&nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 训练就会产生。具体路径和你的代码有关系。代码有一个路径。我没有复现这个，只是看了看。

---原始邮件--- 发件人: "yy835055664"<[email protected]&amp;gt; 发送时间: 2019年12月10日(星期二) 下午5:08 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&amp;gt;; 抄送: "Comment"<[email protected]&amp;gt;;"QI ZHANG"<[email protected]&amp;gt;; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 你好学长： 我想请教你几个问题？ 1.checkpiont里面.h5文件怎么产生的？ 2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？ 望学长能给予解答，谢谢！

------------------&amp;amp;nbsp;原始邮件&amp;amp;nbsp;------------------ 发件人:&amp;amp;nbsp;"QI ZHANG"<[email protected]&amp;amp;gt;; 发送时间:&amp;amp;nbsp;2019年12月10日(星期二) 下午5:04 收件人:&amp;amp;nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&amp;amp;gt;; 抄送:&amp;amp;nbsp;"未、、敢伤"<[email protected]&amp;amp;gt;;"Comment"<[email protected]&amp;amp;gt;; 主题:&amp;amp;nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) util只有函数，你要调用它。训练的久不久，还是要看数据，数据你觉得差不多就行

---原始邮件--- 发件人: "yy835055664"<[email protected]&amp;amp;amp;gt; 发送时间: 2019年12月10日(星期二) 下午4:52 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&amp;amp;amp;gt;; 抄送: "Comment"<[email protected]&amp;amp;amp;gt;;"QI ZHANG"<[email protected]&amp;amp;amp;gt;; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果： 17200,1.8035286664962769 17201,0.8399255871772766 17202,0.7849252820014954 17203,0.20637908577919006 ...... 200847,0.0009674201137386262 200848,0.0009672052692621946 200849,0.0009669908322393894 ...... 382718,0.2163151204586029 382719,0.2672955095767975 382720,0.5557892322540283 得到的train_acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17210,0.0,0.9999999999995,0.99999999999998 ...... 211130,0.0,0.9999999999995,0.99999999999998 211140,0.0,0.9999999999995,0.99999999999998 211000,0.32981049562682213,0.1791044776114538,0.9607142857142851 211000,0.282798833819242,0.1739130434777845,0.9728571428571422 ...... 382440,0.03024781341107871,0.9152542372876369,0.9964285714285708 382450,0.008746355685131196,0.9285714285709286,0.9971428571428564 得到的acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17200,0.0,0.9999999999995,0.99999999999998 ...... 210800,0.0,0.9999999999995,0.99999999999998 211000,0.0,0.9999999999995,0.99999999999998 211000,0.3589743589743589,0.12269938650270346,0.9266666666666662 211000,0.358451072736787,0.10714285714243434,0.9743589743589739 ...... 382200,0.0567765567765568,0.7297297297292311,0.9897435897435893 382400,0.13291470434327576,0.6666666666661832,0.9887179487179483 这时候我就kill掉了train.py，然后python test_model.py得到了结果： f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261 请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？ 学长，抱歉打扰，之前的问题已解决。我用完整的数据集train-clean-100进行了python train.py，好像训练得太久了，得到的losses.txt文件结果： 17200,1.8035286664962769 17201,0.8399255871772766 17202,0.7849252820014954 17203,0.20637908577919006 ...... 200847,0.0009674201137386262 200848,0.0009672052692621946 200849,0.0009669908322393894 ...... 382718,0.2163151204586029 382719,0.2672955095767975 382720,0.5557892322540283 得到的train_acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17210,0.0,0.9999999999995,0.99999999999998 ...... 211130,0.0,0.9999999999995,0.99999999999998 211140,0.0,0.9999999999995,0.99999999999998 211000,0.32981049562682213,0.1791044776114538,0.9607142857142851 211000,0.282798833819242,0.1739130434777845,0.9728571428571422 ...... 382440,0.03024781341107871,0.9152542372876369,0.9964285714285708 382450,0.008746355685131196,0.9285714285709286,0.9971428571428564 得到的acc_eer.txt结果： 17200,0.0,0.9999999999995,0.99999999999998 17200,0.0,0.9999999999995,0.99999999999998 ...... 210800,0.0,0.9999999999995,0.99999999999998 211000,0.0,0.9999999999995,0.99999999999998 211000,0.3589743589743589,0.12269938650270346,0.9266666666666662 211000,0.358451072736787,0.10714285714243434,0.9743589743589739 ...... 382200,0.0567765567765568,0.7297297297292311,0.9897435897435893 382400,0.13291470434327576,0.6666666666661832,0.9887179487179483 这时候我就kill掉了train.py，然后python test_model.py得到了结果： f-measure = 0.5833333333328472, true positive rate = 0.5, accuracy = 0.9857142857142851, equal error rate = 0.08017492711370261 请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？ 你好学长，我看了你复现那个Deep-speaker，也出结果了，我想问你几个问题？ 1.checkpiont里面.h5文件怎么产生的？ 2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？ 望学长能给予解答，谢谢！ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[LittleMaWen]

请问为什么我用编号为1-3的人去训练，然后再用编号为4-5的人去测试，得到的结果却是f-measure = 0.9999999999995, true positive rate = 1.0, accuracy = 0.99999999999998, equal error rate = 0.0，这个结果不就代表着测试的人和训练的人是一样的吗？但实际上并不同啊？你们会这样吗？

[Walleclipse]

请问学长，我是不是训练得太久了？然后这个结果是这样吗？另外请问demo里面的两张图EER.png和loss.png是怎么得到的？我python utils.py好像什么都没得到？

是不是训练太久，你得看一下你的 EER.png 和 Loss.png 如果到后面下降了，就是训练太久了。我觉得你的结果可能没问题。Readme 里的图通过调用 utils.py 里面的 def plot_loss 来画 e.g.

import constants as c
from utils import plot_loss,plot_acc
loss_file=c.CHECKPOINT_FOLDER+'/losses.txt' # loss log file path
plot_loss(loss_file)
acc_file=c.CHECKPOINT_FOLDER+'/acc_eer.txt'# acc log file path
plot_acc(acc_file)

[Walleclipse]

你好学长： 我想请教你几个问题？ 1.checkpiont里面.h5文件怎么产生的？ 2.下载的测试集是直接放到Libri文件里面吗，跟训练语音一块。测试程序需要改动吗？还是训练完了直接就可以测试了？ 望学长能给予解答，谢谢！

1. 运行 train.py 以后每过 SAVE_PER_EPOCHS 步就会保存 checkpiont 里面.h5文件。默认情况下SAVE_PER_EPOCHS=200， 你可以在constants.py 里修改 SAVE_PER_EPOCHS。
2. 测试集放在哪里都可以，但是你需要修改一些路径。假设你把测试集的 wav文件放在了 "audio/LibriSpeechSamples/test-clean-100" ，需要把提其特征后的测试集放在“audio/LibriSpeechSamples/test-clean-100-npy/” 那么：
   1） 在 constants.py 中，设置第三行 TEST_DIR = 'audio/LibriSpeechSamples/test-clean-100-npy/'
   2）调用 pre_process.py 处理测试集

import constants as c
from pre_process import preprocess_and_save
test_wav_dir = "audio/LibriSpeechSamples/test-clean-100"
preprocess_and_save(wav_dir=test_wav_dir,out_dir=c.TEST_DIR )

train.py 就会在训练的过程中测试

[Walleclipse]

请问为什么我用编号为1-3的人去训练，然后再用编号为4-5的人去测试，得到的结果却是f-measure = 0.9999999999995, true positive rate = 1.0, accuracy = 0.99999999999998, equal error rate = 0.0，这个结果不就代表着测试的人和训练的人是一样的吗？但实际上并不同啊？你们会这样吗？

数据太少了，不能说明问题。还是建议下载完整的LibriSpeech数据再运行程序。

[yaoyao1206]

师兄用pre_process.py 预处理数据只能处理wav格式的数据吗，我下载了LibriSpeech数据集里面全是flac格式的，是要先把它转换成wav后再运行pre_process.py吗

[yy835055664]

是的，先把falc格式转换为wav，在进行数据预处理

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "Walleclipse/Deep_Speaker-speaker_recognition_system" <[email protected]&gt;; 发送时间:&nbsp;2020年11月25日(星期三) 下午3:38 收件人:&nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送:&nbsp;"未、、敢伤"<[email protected]&gt;;"Comment"<[email protected]&gt;; 主题:&nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 师兄用pre_process.py 预处理数据只能处理wav格式的数据吗，我下载了LibriSpeech数据集里面全是flac格式的，是要先把它转换成wav后再运行pre_process.py吗 — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[izhangy]

是的，需要先转换下数据格式 &nbsp; Hardin [email protected] &nbsp;

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "yy835055664"<[email protected]&gt;; 发送时间: 2020年11月25日(星期三) 下午3:45 收件人: "Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&gt;; 抄送: ""<[email protected]&gt;; "Comment"<[email protected]&gt;; 主题: Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 是的，先把falc格式转换为wav，在进行数据预处理

------------------&amp;nbsp;原始邮件&amp;nbsp;------------------ 发件人: "Walleclipse/Deep_Speaker-speaker_recognition_system" <[email protected]&amp;gt;; 发送时间:&amp;nbsp;2020年11月25日(星期三) 下午3:38 收件人:&amp;nbsp;"Walleclipse/Deep_Speaker-speaker_recognition_system"<[email protected]&amp;gt;; 抄送:&amp;nbsp;"未、、敢伤"<[email protected]&amp;gt;;"Comment"<[email protected]&amp;gt;; 主题:&amp;nbsp;Re: [Walleclipse/Deep_Speaker-speaker_recognition_system] 运行报错！求助 (#39) 师兄用pre_process.py 预处理数据只能处理wav格式的数据吗，我下载了LibriSpeech数据集里面全是flac格式的，是要先把它转换成wav后再运行pre_process.py吗 — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.