xiph / rnnoise

Recurrent neural network for audio noise reduction

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

UnboundLocalError: local variable 'batch_index' referenced before assignment

ifishtoo opened this issue · comments

I ran
python3 ./bin2hdf5.py ../src/trainning.f32 0 87 training.h5
at the suggestion of the numpy git maintainer. It seemed to work.
I followed that with
$ python3 ./rnn_train.py
py
Using Theano backend.
Build model...
Loading data...
done.
0 sequences
0 train sequences. x shape = (0, 2000, 42) y shape = (0, 2000, 22)
Train...
Train on 0 samples, validate on 0 samples
Epoch 1/120
Traceback (most recent call last):
File "/home/victor/Downloads/git-rnnoise/rnnoise/training/./rnn_train.py", line 112, in
model.fit(x_train, [y_train, vad_train],
File "/usr/lib/python3/dist-packages/keras/engine/training.py", line 1227, in fit
return training_arrays.fit_loop(self, fit_function, fit_inputs,
File "/usr/lib/python3/dist-packages/keras/engine/training_arrays.py", line 205, in fit_loop
if batch_index == len(batches) - 1: # Last batch.
UnboundLocalError: local variable 'batch_index' referenced before assignment

I installed python3 in ubuntu 21.04 via apt and added the proper modules as the error messages indicated. Is there another module I can add to compensate for this error?
FYI: My original .raw files are exported from audacity. They are only about 1 minute each of speech (with noise) and 1 minute of noise. Do I need larger samples? I am just trying to filter out the hum from my sound card. Later I will offload the processing to via opencl.