Working training in colab but no sound

Question

Working training in colab but no sound

pabloriera opened this issue 4 years ago · 7 comments

Hi, I manage to setup a Colab for trainging. The trainng occurs, at least the first 100 generated samples do not have sound just clicks. Do you know how many epochs should take? Or maybe is something uncompatible with versions o other thing wrong going on

https://colab.research.google.com/drive/1fRhzNtRmdllD74mLzfyCy8SWuMT7sB3m

Best

Answer 1 · 2020-08-10T18:38:09.000Z

For me all the Userwarnings were worth fixing. For the sample generation, I have to admit that I didn't succeed to generate as much good sounding sample as the Theano implementation (link here https://github.com/soroushmehr/sampleRNN_ICLR2017).

Moreover, 14mins might be short for your Voice Dataset I think you should look for those 10 hours video. Also, when listening to the YouTube video, I noticed many long periods of time with no sound and especially no human voice. You should consider another dataset like a long podcast or something like that.

I'm sorry for not having a "real" answer, but I hope this could help a bit.

Berthié

Answer 2 · 2020-08-10T20:16:08.000Z

Ok, thanks for the info. I would try to fix the warnings, don't now how.

First I tried with the pianojazz video that lasts 3 hours, but I got the same results, so I switched to a short audio. I was hopping the model to overfitt to the audio at least.

Answer 3 · 2020-09-24T06:05:10.000Z

Hi pabloriera, have you fix your problem?

I also tried training the model using 2 hours' music from a band, following the Readme, but without success.
Then I tried to reproduce the piano example exactly the same as the Readme, but I still got nothing but clips.

I was using a local server with tpu. I also wonder what is going wrong.

Answer 4 · 2020-09-24T15:46:38.000Z

Hi, I could not work it out. I found a keras implementation and manage to have acceptable results. El jue., 24 sept. 2020 a las 3:05, Zhe Zhang (<notifications@github.com>) escribió:

…

Hi pabloriera, have you fix your problem? I also tried training the model using 2 hours' music from a band, following the Readme, but without success. Then I tried to reproduce the piano example exactly the same as the Readme, but I still got nothing but clips. I was using a local server with tpu. I also wonder what is going wrong. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#40 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAMJ7PNCUAKV2HBLAMURM2LSHLOSHANCNFSM4PZOFDCA> .

-- ♫

Answer 5 · 2020-12-23T22:46:46.000Z

Hi! Did anyone have any success implementing this notebook in a local environment?

Answer 6 · 2021-03-08T09:05:15.000Z

285 epochs and still nothing but pops :(

Answer 7 · 2021-12-09T00:24:06.000Z

Hi, I tried to set it up in the local env, but for a small set of audio data, it samples out click noises and nothing else. Anybody got some good results, please share them here.
BR