not a bug: more a feedback with windows machine to be used by others

[alain2208]

Hi i discovered this git few days ago, seems very nice job!

I work on windows and not a great pyhton guy but very interested in voice.

For now the main issues i found are related to path. Python windows don't like a lot relative path and so for now i try to 'make it work' in rewriting path in full not in relative. Also have to add in the environment variable the path to ffmpeg that i have to grab for librosa to work with mp3.

[drscotthawley]

@alain2208 Ok, cool! Send me a PR if you get it working. Preferably the new code could be wrapped in code that tests for the OS. ;-)

[f00f]

I got panotti running on Windows without changing relative paths (but maybe that has been merged in the meantime). I had python 3.6.8, Keras 2.3.1and tensorflow-gpu 2.0.0 installed an running already. Note that I only got panotti running and have not evaluated its performance, yet.

• clone the panotti repo
• install sox, add to PATH
• download ffmpeg, add to PATH

Make sure to open new PowerShell and git bash windows after changing PATH.

Edit ./train_network.py and ./preprocess_data.py: change first lines to

#! /usr/bin/env python

(i.e. python3 -> python)

In PowerShell:

pip install librosa==0.6.3
pip install cython matplotlib imageio pandas audioread h5py librosa Pillow pypandoc scikit_learn winshell scikit-image

The recent libRosa crashed somewhere in binaural_setup.sh.

In git bash:

cd examples
cd ./binaural_setup.sh

Tensorflow complains about dropout values > 0.5:

WARNING:tensorflow:Large dropout rate: 0.6 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.

so I guess this needs to be changed to 0.4.

Hope this helps someone.

[drscotthawley]

Thanks for sharing this! I'll leave this up for others...

[drscotthawley]

Since this is solved, I'll close. They can still find it in searching closed issues.