dataset_gen.py

import numpy as np
import os
import librosa
import librosa.display
import spiegelib
import matplotlib.pyplot as plt

#create class for mel featres
class melSpectrogramFeatures(spiegelib.features.features_base.FeaturesBase):

    def get_features(self, audio):
        features = librosa.feature.melspectrogram(y=audio.get_audio(), sr=self.sample_rate,)
        return features

#create class for raw audio
class audioFeatures(spiegelib.features.features_base.FeaturesBase):

    def get_features(self, audio):
        return audio.get_audio()


#load synth to use
synth = spiegelib.synth.SynthVST("/Library/Audio/Plug-Ins/Components/Serum.component")

#make class for extracting mel spectrogram features
melSpec = melSpectrogramFeatures()

#class instance for getting audio features
audioFet = audioFeatures()

#setup location for dataset generation
output_location = "/Volumes/USB30FD/Synth_DataSet"

#setup generator
generator = spiegelib.DatasetGenerator(synth,melSpec,output_folder=".",save_audio=True)
generator.generate(2, file_prefix="mel_")
#generate audio
generator.generate(3000, file_prefix="train_1_")
generator.generate(3000, file_prefix="train_2_")
generator.generate(3000, file_prefix="train_3_")
generator.generate(2000, file_prefix="test_")