CVxTz
Value Error with input
I am trying to get keras_cnn_mel.py to run but am getting the following error:
ValueError: Error when checking input: expected input_1 to have shape (63, 320, 1) but got array with shape (157, 320, 1)
I’m guessing the time dimension is not matching up for some of the audio files. Did you end up truncating those files or is there some sort of data compression technique I should look into?
Hello All your audio files should have the same length, so you need to truncate them or pad them. Best
what padding technique do you recommend? or would you be able to share the code you used to produce the results in the medium post?
Its in the repo : https://github.com/CVxTz/audio_classification/blob/master/code/keras_cnn_mel.py
def load_audio_file(file_path, input_length=input_length): data = librosa.core.load(file_path, sr=16000)[0] #, sr=16000 if len(data)>input_length:
max_offset = len(data)-input_length
offset = np.random.randint(max_offset)
data = data[offset:(input_length+offset)] # HERE: Truncating
else:
if input_length > len(data):
max_offset = input_length - len(data)
offset = np.random.randint(max_offset)
else:
offset = 0
data = np.pad(data, (offset, input_length - len(data) - offset), "constant") ##HERE : Padding
data = preprocess_audio_mel_T(data)
return data
Hmm, I haven't changed that part though, so the data is being loaded by the load_audio_file code. It seems like the error actually doesn't have to do with the data being input into the model, but potentially the data being output by the first layer (input layer) of the model. Any idea what could be happening? I haven't changed the code except to download the data and put that file in my directory.
Maybe its a difference in librosa version since I uploaded the code Maybe you can change the input shape from " inp = Input(shape=(63, 320, 1))" to " inp = Input(shape=(157, 320, 1))" in the model definition ?
That also did not work, and I am getting the same error message. Could you let me know what versions of each package you used in your working version of the code?