The loss can not decrese

[zyy-fc]

Thank you for sharing!
I used the 4620 utterances from the training set of the TIMIT corpus as train set, and 119 types noise as train set. Besides, the 10 utterances from the test set of the TIMIT corpus and other 10 types noise as validation set. I have trained this model for more than 10 hours, but the loss seems to be invariant.
I do not know why. I will be honoured if you help me.

[zhr1201]

I think it would be better if you could use some kind of plot (e.g. tensorboard ) to visualize the convergence and test some samples, also try to do some parameter tunning. If the model still couldn't generate satisfactory samples, probably adding more layers and use more advance structures may help.

[zyy-fc]

than you very much!

[zhr1201]

Not sure what the problem is, I guess it is very likely becaues of version issues. You can set a stop at stft to see what is going wrong. On 11/07/2018 11:12, yuanyuan0209 wrote: But I am also having another problem and want to ask if you solved it. I have tried different approaches but all to no available . — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[bailiangze]

it occurs when i use your code .it likes below
Exception in thread Thread-7:
Traceback (most recent call last):
File "E:\anaconda\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "E:\anaconda\lib\threading.py", line 864, in run
self._target(*self._args, **self.kwargs)
File "D:\deeplearning\CNN-for-single-channel-speech-enhancement-master\audio_reader.py", line 108, in thread_main
noise_org, _ = librosa.load(self.noisefiles[noise_id], sr=None)
File "E:\anaconda\lib\site-packages\librosa\core\audio.py", line 112, in load
with audioread.audio_open(os.path.realpath(path)) as input_file:
File "E:\anaconda\lib\site-packages\audioread_init.py", line 116, in audio_open
raise NoBackendError()
audioread.NoBackendError

but i test the code l'ibrosa; it does not have problem。can you give me some advice

[ffxz]

@yuanyuan0209 is your loss is decreased? can your share your successful results?

[lvyilan23]

@yuanyuan0209 is your loss is decreased? can your share your successful results?

hi, have you solve the problem? I meet the same question

[ZBang]

I meet the same problem.
if you hava some ideas,please tell me
thanks!

[lvyilan23]

i have no idea, so I changed the code to reproduce the model 发自我的 iPad

…

------------------ Original ------------------ From: ZBang <notifications@github.com&gt; Date: Tue,Jun 16,2020 1:39 PM To: zhr1201/CNN-for-single-channel-speech-enhancement <CNN-for-single-channel-speech-enhancement@noreply.github.com&gt; Cc: lvyilan23 <254467442@qq.com&gt;, Comment <comment@noreply.github.com&gt; Subject: Re: [zhr1201/CNN-for-single-channel-speech-enhancement] The loss can not decrese (#9) I meet the same problem. if you hava some ideas,please tell me thanks! — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.