fedden

import os
import librosa

directory = './path/to/my/audio/folder/'

for file in os.listdir(directory):
    if file.endswith('.wav'):
        file_path = os.path.join(directory, file)
        audio_data, _ = librosa.load(file_path)

fedden

from magenta.models.nsynth import utils
from magenta.models.nsynth.wavenet import fastgen

def wavenet_encode(file_path):
    
    # Load the model weights.
    checkpoint_path = './wavenet-ckpt/model.ckpt-200000'
    
    # Load and downsample the audio.
    neural_sample_rate = 16000

fedden

import librosa

sample_rate = 44100
mfcc_size = 13

# Load the audio
pcm_data, _ = librosa.load(file_path)

# Compute a vector of n * 13 mfccs
mfccs = librosa.feature.mfcc(pcm_data, 

fedden

ls -lah ./audio_dataset/

...
-rw-rw-r-- 1 tollie tollie  3.8M Jun 28  2014 HAL9K - Long Sustained Note.wav
-rw-rw-r-- 1 tollie tollie  2.7M Jul  2  2014 HAL9K - Lost Soul.wav
-rw-rw-r-- 1 tollie tollie  7.5M Jun 29  2014 HAL9K - Low Long Tail.wav
-rw-rw-r-- 1 tollie tollie  3.8M Jun 28  2014 HAL9K - Low Short.wav
-rw-rw-r-- 1 tollie tollie  4.6M Jun 28  2014 HAL9K - Low Thump.wav
-rw-rw-r-- 1 tollie tollie  4.6M Jul  2  2014 HAL9K - Lute 1.wav
-rw-rw-r-- 1 tollie tollie  7.7M Jul  2  2014 HAL9K - Lute 2.wav

fedden

import numpy as np

# Create some random MFCC shaped features as a sequence of 10 values
feature_sequence = np.random.random((10, 13))

# Get the standard deviation
stddev_features = np.std(feature_sequence, axis=0)

# Get the mean
mean_features = np.mean(feature_sequence, axis=0)

fedden

from sklearn.decomposition import PCA
from sklearn.preprocessing import MinMaxScaler


def get_pca(features):
    pca = PCA(n_components=2)
    transformed = pca.fit(features).transform(features)
    scaler = MinMaxScaler()
    scaler.fit(transformed)
    return scaler.transform(transformed)

fedden

import umap
from sklearn.preprocessing import MinMaxScaler


def get_scaled_umap_embeddings(features, neighbour, distance):
    
    embedding = umap.UMAP(n_neighbors=neighbour,
                          min_dist=distance,
                          metric='correlation').fit_transform(features)
    scaler = MinMaxScaler()

fedden

from sklearn.manifold import TSNE
from sklearn.preprocessing import MinMaxScaler


def get_scaled_tsne_embeddings(features, perplexity, iteration):
    embedding = TSNE(n_components=2,
                     perplexity=perplexity,
                     n_iter=iteration).fit_transform(features)
    scaler = MinMaxScaler()
    scaler.fit(embedding)

fedden

fedden