imfing / audio-classification

:musical_score: Environmental sound classification using Deep Learning with extracted features

Hi there 👋

LinkedIn | Homepage | Twitter

👨‍💻 Working on AI platform

Fun Facts

• Fing is my nickname
• I used mtobeiyf as username which comes from The Hunger Games quote: "May The Odds Be Ever In Your Favor"

audio-classification's Issues

wav file does not work

When I try to put wav files in data directory, it appears:
[Error] extract feature error in data_wav/6/148835-6-0-0.wav. Frequency band exceeds Nyquist. Reduce either fmin or n_bands.
Solution: tune "fmin" parameter in function "extract_feature(file_name=None)" as follows:
contrast = np.mean(librosa.feature.spectral_contrast(S=stft, sr=sample_rate, fmin=180.0).T,axis=0)

Frequency band exceeds Nyquist. Reduce either fmin or n_bands.

I am experiencing the same issue. Should I prep the .wav files in a specific way?

Also, if I reduce fmin to 120.0, I receive the following error:
librosa.util.exceptions.ParameterError: Filter pass-band lies beyond Nyquist

Please advise.

Okay, thanks. It's a problem related to the audio files.
I'll close this issue for now.

Originally posted by @mtobeiyf in #10 (comment)

still getting error after executing the command cnn.py -p

Steps which I did..
1)download ESC-10 dataset then inserted in data directory
2)executed command python feat_extract.py
3)4 file are created feat.py,lable.py,predict_feat,predict_filenames
4)executed command svm.py and output is fitting... acc=0.756
5)executed nn.py
6)executed cnn.py -t
7)putting single sample audio at predict folder and executed the command cnn.py -t but erorr occures like

Traceback (most recent call last):
File "cnn.py", line 123, in
main(args)
File "cnn.py", line 109, in main
elif args.predict: predict(args)
File "cnn.py", line 73, in predict
pred = model.predict_classes(X_predict)
File "C:\Users\Abhijeet\Anaconda3\envs\tfp3.6\lib\site-packages\keras\engine\s
equential.py", line 268, in predict_classes
if proba.shape[-1] > 1:
AttributeError: 'list' object has no attribute 'shape'

errors in cnn.py

I got two errors when executing the file "cnn.py".

1. Unresolved reference 'feat_extract' when we define the function:train(args).
   solution for this error: we should use "import feat_extract" instead of using "from feat_extract import *";
   2.Unresolved reference 'X_predict' when we define the function: predict(args).
   "X_predict = np.expand_dims(X_predict, axis=2)", the "X_predict" in the right side is not pre-defined nor pre-produced.
   solution for this error: we can replace "X_predict" by "predict_feat_path".

The class labels seem to differ from the actual labels.

When I tried to predict the classes there seemed to be some kind of pattern. The class names given by folder is different from the predicted name:
Name, Predicted,Actual
rain=0,2
seawaves=1,3
baby=2,4
clock=3,5
sneeze=4,6
helicopter=5,7
chainsaw=6,8
rooster=7,9
firecracker=8,10
dog=13,1
What could be the reason?

steps for how to create custom model to run cnn.py -p command ????/

index out of bound

when executing nn.py gives following error

Using TensorFlow backend.
Traceback (most recent call last):
File "nn.py", line 37, in
y_train = keras.utils.to_categorical(y_train-1, num_classes=10)
File "C:\Users\TUSHAR\Anaconda3\lib\site-packages\keras\utils\np_utils.py", line 32, in to_categorical
categorical[np.arange(n), y] = 1
IndexError: index 21 is out of bounds for axis 1 with size 10

error on svm.ph

I got the following error after running svm.py. How can I solve this one?

error ocuures when python cnn.py -p is executed like list object has no attribute 'shape''

I want the versions of these packages

numpy
librosa
pysoundfile
sounddevice
matplotlib
scikit-learn
tensorflow
keras

how to get .ogg file

local variable 'batch_index' referenced before assignment

I have run feat_extract.py，and there is feat.npy and ;label.npy,when I run cnn.py,it occured error above

Error! feat.npy is not UTF-8 encoded

Hi , i am using the same structure as you said but when i run feat_extract.py ,it run smoothly and gives me this output :

extract .ipynb_checkpoints features done
extract 001 - Dog bark features done
extract 002 - Rain features done
extract 003 - Sea waves features done
extract 004 - Baby cry features done
extract 005 - Clock tick features done
extract 006 - Person sneeze features done
extract 007 - Helicopter features done
extract 008 - Chainsaw features done
extract 009 - Rooster features done
extract 010 - Fire crackling features done

but in feat.npy, it says Error! feat.npy is not UTF-8 encoded

please help me to understand this and how i can resolve it

what accuracy can this cnn model reach?

what's the optimal accuracy can this cnn model reach after fully trained, can this be written on README ? thanks

prediction

Can you guide the prediction part?

Can I use .wav to train instead of .ogg

0 test accuracy

conda env update -n my_env --file ENV.yaml gives 'utf8' codec can't decode byte 0x88 in position 2

I get an index 12 out of bounds when I run the cnn.py -t

So this happened with both my own OGG data and the ESC data, with one OGG file in the predict folder.

Everything went normally, the SVN and MLP worked. Then I tried the CNN:

_

Phonecian:audio-classification skiwheelr$ python3 cnn.py -t
Using TensorFlow backend.
2020-04-03 23:29:32.002190: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-04-03 23:29:32.015288: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x7f8e13dfbac0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-04-03 23:29:32.015310: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version
Traceback (most recent call last):
File "cnn.py", line 124, in
main(args)
File "cnn.py", line 109, in main
if args.train: train(args)
File "cnn.py", line 51, in train
y_train = keras.utils.to_categorical(y_train, num_classes=class_count)
File "/usr/local/lib/python3.7/site-packages/keras/utils/np_utils.py", line 52, in to_categorical
categorical[np.arange(n), y] = 1
IndexError: index 12 is out of bounds for axis 1 with size 12

What could be causing this? There is enough data.

UnboundLocalError: local variable 'X_predict' referenced before assignment

The following is the error I have encountered. Could you please guide me out of it?

Traceback (most recent call last):
File "cnn.py", line 139, in
main(args)
File "cnn.py", line 125, in main
elif args.predict: predict(args)
File "cnn.py", line 88, in predict
X_predict = np.expand_dims(X_predict, axis=2)
UnboundLocalError: local variable 'X_predict' referenced before assignment

Attribute error

Hello, when I run nn.py and cnn.py, the spyder showed that AttributeError: 'ProgbarLogger' object has no attribute 'log_values' . It seems that something wrong the keras

keras.utils.to_categorical(y, num_classes) question

Hello,
I have a question.

nn.py (under 34 line)

Convert label to onehot
y_train = keras.utils.to_categorical(y_train-1, num_classes=num_classes)
y_test = keras.utils.to_categorical(y_test-1, num_classes=num_classes)

Why did you use (y_train - 1), not just (y_train) when you call keras.utils.to_categorical()

Classes

What are the classes if I use ECS-10? Is it like: 0 - dog bark, 1 - ***, ..., etc.

I put a dog bark.ogg from the training data into the predict folder, it outputs 13 however. What is wrong with it?

Thank you.