Loading Data - Deep Learning for Audio Classification p.2

Data can be found at this Github repo:

Kaggle competition where data is taken from:

[seth]

I have re-worked this series with the full code on github. If you want to follow along from these videos you still can. After cloning your repo:

git checkout

Will revert to the most recent commit before I overhauled everything.

[shafagh_projects]

because the wav file are in their corresponding folder I think the code needed to be adjusted as follow:


df.set_index('filename', inplace=True)

for f in df.index:
label = df.at[f, 'label'] # Get the label for the file
filepath = f'wavfiles/{label}/{f}' # Construct the file path
rate, signal = wavfile.read(filepath) # Read the WAV file
df.at[f, 'length'] = signal.shape[0] / rate # Calculate and set the length

[wuzark]

One million thanks for creating such an interesting and very well explained content!!!

[droy]

The background low frequency bass sound is really distracting in this video maybe do a high pass filter on your audio. The issue happens when you are typing. However, great tutorial so far.

[Niculwmusic]

You can do librosa.load(file, sr=None) to return the original sampling rate

[KordTaylor]

Thank you. Another great video and really practical with the files and all. 👏🏻🏅🔥

[jugsma]

under the iteration of classes, line 70:
instead of using librosa to get signals can't we use wavefile.read().
I notice the difference, however, what would be the difference if we:
Instead of : librosa.load('wavfiles/'+wav_file, sr=44100)
we use: np.array([x/rate for x in

[jeremynx]

thank you, awesome explanation <3
I am your fan now!!!

[MathStatsMe]

When I try to load my particular dataset with librosa, I get the error that I can only concatenate str (not "numpy.int64") to str. I've checked that the way I've defined wav_file is a string, but don't understand how the folder name wouldn't be a string. Any ideas here? I have a folder in .venv called "All WAV files". Here is my code:

signal, rate = librosa.load('All WAV files/'+wav_file, sr=1000)

[lucasgrassoramos]

Hey, great video! where can i find the .csv file?

[bouchrabouchra]

hi, i'm having an issue with this line "signal, rate = files\audio_train/' +wav_file, sr=44100 )", and this is the error that pops up when i execute it
RuntimeError Traceback (most recent call last)
in load(path, sr, mono, offset, duration, dtype, res_type)
128 try:
--> 129 with sf.SoundFile(path) as sf_desc:
130 sr_native = sf_desc.samplerate

in __init__(self, file, mode, samplerate, channels, subtype, endian, format, closefd)
628 format, subtype, endian)
--> 629 self._file = self._open(file, mode_int, closefd)
630 if set(mode).issuperset('r+') and self.seekable():

in _open(self, file, mode_int, closefd)
1183 _error_check(_snd.sf_error(file_ptr),
-> 1184 "Error opening {0!r}: ".format(self.name))
1185 if mode_int == _snd.SFM_WRITE:

in _error_check(err, prefix)
1356 err_str = _snd.sf_error_number(err)
-> 1357 raise RuntimeError(prefix + _ffi.string(err_str).decode('utf-8', 'replace'))
1358

RuntimeError: Error opening files\\audio_train/Writing': System error.

During handling of the above exception, another exception occurred:

FileNotFoundError Traceback (most recent call last)
in <module>
----> 1 signal, rate = files\audio_train/' +wav_file, sr=44100 )

in load(path, sr, mono, offset, duration, dtype, res_type)
145 if isinstance(path, six.string_types):
146 warnings.warn('PySoundFile failed. Trying audioread instead.')
--> 147 y, sr_native = __audioread_load(path, offset, duration, dtype)
148 else:
149 six.reraise(*sys.exc_info())

in __audioread_load(path, offset, duration, dtype)
169
170 y = []
--> 171 with audioread.audio_open(path) as input_file:
172 sr_native = input_file.samplerate
173 n_channels = input_file.channels

in audio_open(path, backends)
109 for BackendClass in backends:
110 try:
--> 111 return BackendClass(path)
112 except DecodeError:
113 pass

in __init__(self, filename)
60 """
61 def __init__(self, filename):
---> 62 self._fh = open(filename, 'rb')
63
64 try:

FileNotFoundError: [Errno 2] No such file or directory: files\\audio_train/Writing'

RuntimeError Traceback (most recent call last)
in load(path, sr, mono, offset, duration, dtype, res_type)
128 try:
--> 129 with sf.SoundFile(path) as sf_desc:
130 sr_native = sf_desc.samplerate

in __init__(self, file, mode, samplerate, channels, subtype, endian, format, closefd)
628 format, subtype, endian)
--> 629 self._file = self._open(file, mode_int, closefd)
630 if set(mode).issuperset('r+') and self.seekable():

in _open(self, file, mode_int, closefd)
1183 _error_check(_snd.sf_error(file_ptr),
-> 1184 "Error opening {0!r}: ".format(self.name))
1185 if mode_int == _snd.SFM_WRITE:

in _error_check(err, prefix)
1356 err_str = _snd.sf_error_number(err)
-> 1357 raise RuntimeError(prefix + _ffi.string(err_str).decode('utf-8', 'replace'))
1358

RuntimeError: Error opening files\\audio_train/Writing': System error.

During handling of the above exception, another exception occurred:

FileNotFoundError Traceback (most recent call last)
in <module>
----> 1 signal, rate = files\audio_train/' +wav_file, sr=44100 )

in load(path, sr, mono, offset, duration, dtype, res_type)
145 if isinstance(path, six.string_types):
146 warnings.warn('PySoundFile failed. Trying audioread instead.')
--> 147 y, sr_native = __audioread_load(path, offset, duration, dtype)
148 else:
149 six.reraise(*sys.exc_info())

in __audioread_load(path, offset, duration, dtype)
169
170 y = []
--> 171 with audioread.audio_open(path) as input_file:
172 sr_native = input_file.samplerate
173 n_channels = input_file.channels

in audio_open(path, backends)
109 for BackendClass in backends:
110 try:
--> 111 return BackendClass(path)
112 except DecodeError:
113 pass

in __init__(self, filename)
60 """
61 def __init__(self, filename):
---> 62 self._fh = open(filename, 'rb')
63
64 try:

FileNotFoundError: [Errno 2] No such file or directory: files\\audio_train/Writing'

please help

[jeremynx]

when I run your code with reading length of audio file, I got NaN values for several labels. I think there is something wrong. please help

[oscarzuniga]

Hello, u know how i can solve the problem that my audiodata isn't in 16 bits?

[rekhapandey]

Hey can u suggest audio filtering techniques
Like I have two files one is noise file
Second is noise + signal
How to remove this

[SHADABALAM]

Hi Seth.. loving ur this series.. thumbs up... i have a question as a newbie ... my data set consists of audio file of 17second each file and about 500 files are in my data set of 3 categories. Is this length i.e. 17 second is normal for creating spectrogram or mfcc?? all files contains full audio from beginning to end. cant perform any cleaning operation on them. TIA.

[zhannafedorova]

Hello!
Thank you for your videos, they're really helpful!

Could you please explain why you don't use librosa library functions to calculate mel_spectr? Is it not optimized or something else? I don't know what to chose in my project.

[upsbmussy]

Thank you so much, I'm using this for my science project, and would like to ask if this tutorial would be effective to detect notes?

[lmor]

Man, I've really liked your video, thank you very much!

[renarrow]

Hey Seth.i'm a bit of a beginner. loving these videos, well done. Having an issue at df.set_index('fname', inplace=True) . This error pops up -raise KeyError("None of {} are in the columns".format(missing))
KeyError: "None of ['fname'] are in the columns" Cant seem to find a solution. Kindly help me out.

[ahmaddaffa]

Hello I want to ask where I can get the file of instrument.csv because I already check from you latest video and github there's no such as file instrument.csv. Thank you