Error Extracting MFCCs: Could someone please lend a helping hand

[jamila Muhammad]

import json

import math

import librosa

import os

DATASET_PATH = '/content/drive/MyDrive/ADReSS/ADReSS2020/ADReSS-IS2020-Train-data/train/Full_wave_enhanced_audio'

JSON_PATH = '/content/drive/MyDrive/ADReSS/ADReSS2020'

SAMPLES_TO_CONSIDER = 22050 #1 sec worth of sound

SAMPLE_RATE = 22050

DURATION = 30 #measured in seconds

SAMPLES_PER_TRACK = SAMPLE_RATE * DURATION

def save_mfcc(dataset_path, json_path,n_mfcc=13, hop_lenth=512, n_fft=2048, num_segments= 10):

 

  # Get the folder name

  folder_name = "json_files"

  # Check if the folder exists

  if not os.path.exists(os.path.join(JSON_PATH, folder_name)):

    # Create the folder

    os.mkdir(os.path.join(JSON_PATH, folder_name))

 #dictionary to store data

  data= {

        "mappings": [],

        "labels": [],

        "MFCCs": [],

        "files":[]

        }

  num_samples_per_segments = int(SAMPLES_PER_TRACK / num_segments)

  expected_num_mfcc_vectors_per_segment = math.ceil(num_samples_per_segments / hop_lenth) #1.2 -> 2

  #loop through all the sub-dirs

  for i, (dirpath, dirnames, filenames) in enumerate(os.walk(dataset_path)):

    #we need to ensure that we're not at the root level

    if dirpath is not dataset_path:

      #save the semantic level

      dirpath_components = dirpath.split("/") #dataset_class/ Alzheimers

      semantic_label = dirpath_components[-1]

      data["mappings"].append(semantic_label)

      print("\nProcessing{}".format(semantic_label))

      #process files for a specific class

      for f in filenames:

        #load audio file

        filepath = os.path.join(dirpath, f)

        signal , sr = librosa.load(filepath, sr=SAMPLE_RATE)

        #process segments extracting mfcc and storing data

        for s in range(num_segments):

         

          start_sample= num_samples_per_segments * s # s=0 -> 0

          finish_sample = start_sample + num_samples_per_segments # s=0 num_samples_per_segment

         

         

          #extract mfcc

         

          mfcc= librosa.feature.mfcc(signal[start_sample:finish_sample], sr =sr,

                                       n_fft = n_fft,

                                       n_mfcc = n_mfcc,

                                       hop_lenth = hop_lenth)

                                       

          mfcc = mfcc.transpose

           #store mfcc for segment if it has expected length

          if len(mfcc) == expected_num_mfcc_vectors_per_segment:

            data["mfcc"].append(mfcc.tolist())

            data["labels"].append(i-1)

            print("{}, segment:{}".format(filepath, s+1))

  with open(json_path, "w") as fp:

    json.dump(data, fp, indent = 4)

if __name__ == "__main__":

  save_mfcc(DATASET_PATH, JSON_PATH, num_segments=10)

[Pushkar]

Hi Jamila,

Try using keyword arguments instead of positional arguments, i.e

mfcc= librosa.feature.mfcc(y=signal[start_sample:finish_sample], sr =sr,

                                       n_fft = n_fft,

                                       n_mfcc = n_mfcc,

                                       hop_lenth = hop_lenth)

-

Pushkar

--
You received this message because you are subscribed to the Google Groups "librosa" group.
To unsubscribe from this group and stop receiving emails from it, send an email to librosa+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/librosa/1fe1a6b3-8438-4a53-a21c-0099c5d410f0n%40googlegroups.com.

[jamila Muhammad]

Thank you so much for your guide @pushkar, your solution solved it.