Help with


I am rather new to all of this, so please forgive my questions. I am very interested in using the repository at to train a model with 33 hours of audio that I have put into a single .wav file (that I’m unsure what exactly to do with), and produce AI generated music. I would be interested in adding to my 33 hour .wav file for future training sessions. In the meantime, I believe I have been successful in installing all of the dependencies on my CUDA enabled device (Windows 10, Intel Core i7 - 16GB Memory - NVIDIA GeForce RTX 2070 - 512GB SSD + 1TB+8GB Hybrid Hard Drive , Alienware 17 laptop) but when i try to run the command, I get an invalid syntax error at line 25

Experiment will start at:"


I’ve attempted to see if I can change the code in but when I ran a test of the code without the “:” I still received the same error on the same line, still under quotation mark. Not sure if that’s even the right way to go about this.

Anyway, I could use more clarificatoin with the “Training” section of the repository to better understand this process. I feel like I don’t see when I’m supposed to enter specific parameters for the output files I am hoping to generate, or where I even designate the 33-hour long wave file as the dataset (I’ve named it preprocess_all_audio.wav so that I don’t have to actually change much of the code itself), and I imagine missing where I’m supposed to input values (such as the length of this 33 hour, 33 minute, 52 second .wav file) could easily result in invalid syntax errors. FYI I have python37 in my path. 

Would appreciate any help anyone had to offer

Thanks and best regards,

EDIT: The error above was due to the fact the script was written for Python 2.6 not Python 3. The errors in this post have been resolved although the project itself isn’t yet completed, it no longer concerns this orginal link, but a fork of the repository that can be found here: I was able to get my 33 hour file split into 8 second .flac files throuh the two tier method, however it seems I have unrecognized argments in for --n_secs 240 --n_seqs 100

The likeliest thing I can think of being an issue on line 25 is the program is written for Python 2.x. Are you using Python 3.x ?



Thank you for that note, I had played with some workarounds adapting for python 3, but eventually I simply ended up moving on to a different fork of the same code, running it through Ubuntu 64-bit 19.04 as a virtual machine in VMWare. I was able to generate the dataset just fine, and i think i completed the testing section without error, but as for generating the results ive had a couple issues:

  1. can’t connect to gpu from within vmware. The code has device=gpu0, which is outdated and I replaced it to device=cuda9, with the cuda driver installed, and pygpu updated (and even at times, downgraded I believe), but still to no avail. can’t seem to install the nvidia drivers for linux in my virtual machine either (running the driver installer produces an error message: this document was not utf-8 valid)

  2. certain modules aren’t acting as they should, resulting in ‘module’ object has no attribute “print_model_settings” from lib.print_model_settings, or my system can’t import datasets.dataset because no datasets.dataset module exists…

So I’ve made a lot of progress but, things aren’t looking great yet. Appreciate any help anyone has to offer.

EDIT: I’ve even managed to input other scripts from the models/two_tier/ folder and get to a point where I can seemingly debug or resume, but an error message about the “” script brings up unrecognized arguments for --n_secs 240 --n_seqs 100

I get this error message with another script in this folder as well

Best regards,