Closed OnlinePage closed 1 year ago
its a minor bug it will be helpful if you could share your audio files to recreate the error
I had this similar issue, I fixed it by recording the hot word samples for 2 seconds instead of 1 but when i run the inference script it throws a similar error
assert inpAudio.shape == (self.window_frames, ) #1.5 sec long window
AssertionError
i am attaching both json file and my inference code to recreate the error
EfficientWord-Net Version: 1.0.2
Have recently made a commit which should fix the same, however its not up in the pypi repo yet, can you try installing from source using github?
hi @TheSeriousProgrammer thanks, it's working fine now though it often throws this error
OSError: [Errno -9981] Input overflowed
to fix this, I just updated your stream.py in line number 94
to this
get_next_frame = lambda : (
np.frombuffer(mic_stream.read(CHUNK,exception_on_overflow = False),dtype=np.int16)
),
I just passed the exception_on_overflow as False in the pyaudio stream read function
I just wanted to let you know so that you could add it to your next commit, maybe...
that sounds good , can you make a pr of the same? @zerocool-11
Done... :) @TheSeriousProgrammer
Hi been trying to create the new reference file with below command
python -m eff_word_net.generate_reference --input-dir ./mywakewordaudios --output-dir ./ --wakeword mywakeword --model-type resnet_50_arc
Also seems like when generating audio for the same wakeword using generate audio command I was getting 500 response, maybe api request is being limited.
But I created around 8 audios(1-1.5s long) of my wake word using audacity and yet failed to create reference file using resnet_50_arc. However with First_Iteration_Siamese reference file is being easily created.
here is the error trace I got:
eff_word_net\audio_processing.py", line 199, in audioToVector assert inpAudio.shape == (self.window_frames, ) #1.5 sec long window