mix_clips_batch issue

[pi19404]

I am using the function openwakeword.data.mix_clips_batch to mix foreground and background clips at desired snr levels .
And getting the below error .

for batch in tqdm(mixing_generator, total=N_total//batch_size):

File "/workspace/.pyenv_mirror/user/current/lib/python3.8/site-packages/tqdm/std.py", line 1181, in iter
for obj in iterable:
File "/workspace/f1/openwakeword_eval/openwakeword/data.py", line 466, in mix_clips_batch
error_index = torch.from_numpy(np.where(mixed_clips_batch.max(dim=1) != 0)[0])
TypeError: _amax() got an unexpected keyword argument 'dim'

To resolve the issue I had to make changes in data.py file replace the line
error_index = torch.from_numpy(np.where(mixed_clips_batch.max(dim=1) != 0)[0])
with
error_index = torch.from_numpy(np.where(mixed_clips_batch.max(dim=1) != 0)[0])

[dscripka]

I'm not sure I understand the issue, it looks like your corrected line is the same?

Also, while mix_clips_batch still works, the official training process uses the augment_clips function, which should perform a bit better and be more efficient.

[markdewing]

I ran into this issue when running the training_models notebook. The issue is that torch max uses 'dim' as the keyword argument and numpy max uses 'axis'.

The variable 'mixed_clips_batch' is converted to a numpy array here:

openWakeWord/openwakeword/data.py

Line 463 in 7607f95

mixed_clips_batch = (mixed_clips_batch.numpy()*32767).astype(np.int16)

The necessary change in line 466 is

-        error_index = torch.from_numpy(np.where(mixed_clips_batch.max(dim=1) != 0)[0])
+        error_index = torch.from_numpy(np.where(mixed_clips_batch.max(axis=1) != 0)[0])