DatasetGenerator Class

Dataset Generator Class

class spiegel.core.dataset_generator.DatasetGenerator(synth, features=<spiegel.features.mfcc.MFCC object>, outputFolder='/Users/jshier/Development/Academic/spiegel/docs', saveAudio=False, normalize=True)
Parameters

  • synth (Object) – Synthesizer to generate test data from. Must inherit from spiegel.synth.synth_base.SynthBase.

  • features (Object) – Features to use for dataset generation, defaults to spiegel.features.mfcc.MFCC Must inherit from spiegel.features.features_base.FeaturesBase

  • outputFolder (str, optional) – Output folder for dataset, defaults to currect working directory

  • saveAudio (bool, optional) – whether or not to save rendered audio files, defaults to False

  • normalize (bool, optional) – whether or not to normalize features. Requires the normalizers in the feature object to be pre-trained. generate() can be used to train the normalizers. Defaults to True.

Variables

  • featuresFileName (str) – filename for features output file, defaults to features.npy

  • patchesFileName (str) – filename for patches output file, defaults to patches.npy

  • audioFolderName (str) – folder name for the audio output if used. Will be automatically created within the output folder if saving audio. Defaults to audio

createAudioFolder()

Check for and create the audio output folder if necassary

generate(size, filePrefix='', fitNormalizers=False)

Generate dataset with a set of random patches

Parameters

  • size (int) – Number of patches to include in dataset

  • filePrefix (str, optional) – filename prefix for output dataset, defaults to “”

  • fitNormalizers (bool, optional) – Use this dataset to train/fit the normalizers in the feature object. Defaults to False.

saveNormalizers(fileName)

Save feature normalizers

Parameters

fileName (str) – file name for normalizer pickle