- Deskripsi :
NSynth Dataset adalah kumpulan data audio yang berisi ~ 300 ribu not musik, masing-masing dengan nada, warna suara, dan amplop yang unik. Setiap catatan dianotasi dengan tiga informasi tambahan berdasarkan kombinasi evaluasi manusia dan algoritme heuristik: Sumber, Keluarga, dan Kualitas.
Situs web : https://g.co/magenta/nsynth-dataset
Kode sumber :
tfds.audio.Nsynth
Versi :
-
2.3.0
: Fitur baru dalamloudness_db
dalam desibel (tidak dinormalisasi). -
2.3.1
: F0 dihitung dengan perbaikan normalisasi di CREPE. -
2.3.2
: Menggunakan fitur Audio. -
2.3.3
(default): F0 dihitung dengan perbaikan dalam normalisasi gelombang CREPE ( https://github.com/marl/crepe/issues/49 ).
-
Cache otomatis ( dokumentasi ): Tidak
Kunci yang diawasi (Lihat
as_supervised
doc ):None
Kutipan :
@InProceedings{pmlr-v70-engel17a,
title = {Neural Audio Synthesis of Musical Notes with {W}ave{N}et Autoencoders},
author = {Jesse Engel and Cinjon Resnick and Adam Roberts and Sander Dieleman and Mohammad Norouzi and Douglas Eck and Karen Simonyan},
booktitle = {Proceedings of the 34th International Conference on Machine Learning},
pages = {1068--1077},
year = {2017},
editor = {Doina Precup and Yee Whye Teh},
volume = {70},
series = {Proceedings of Machine Learning Research},
address = {International Convention Centre, Sydney, Australia},
month = {06--11 Aug},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v70/engel17a/engel17a.pdf},
url = {http://proceedings.mlr.press/v70/engel17a.html},
}
- Gambar ( tfds.show_examples ): Tidak didukung.
nsynth / full (konfigurasi default)
Deskripsi konfigurasi : Set Data NSynth Lengkap dibagi menjadi set train, valid, dan pengujian, tanpa instrumen yang tumpang tindih antara set train dan set valid / pengujian.
Ukuran unduhan :
73.07 GiB
Ukuran
73.09 GiB
data :73.09 GiB
Split :
Membagi | Contoh |
---|---|
'test' | 4.096 |
'train' | 289.205 |
'valid' | 12.678 |
- Fitur :
FeaturesDict({
'audio': Audio(shape=(64000,), dtype=tf.float32),
'id': tf.string,
'instrument': FeaturesDict({
'family': ClassLabel(shape=(), dtype=tf.int64, num_classes=11),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=1006),
'source': ClassLabel(shape=(), dtype=tf.int64, num_classes=3),
}),
'pitch': ClassLabel(shape=(), dtype=tf.int64, num_classes=128),
'qualities': FeaturesDict({
'bright': tf.bool,
'dark': tf.bool,
'distortion': tf.bool,
'fast_decay': tf.bool,
'long_release': tf.bool,
'multiphonic': tf.bool,
'nonlinear_env': tf.bool,
'percussive': tf.bool,
'reverb': tf.bool,
'tempo-synced': tf.bool,
}),
'velocity': ClassLabel(shape=(), dtype=tf.int64, num_classes=128),
})
- Contoh ( tfds.as_dataframe ):
nsynth / gansynth_subset
Deskripsi konfigurasi : Set Data NSynth terbatas pada instrumen akustik dalam interval nada MIDI [24, 84]. Menggunakan pemisahan alternatif yang memiliki instrumen yang tumpang tindih (tetapi bukan catatan yang tepat) antara set kereta dan set valid / pengujian. Varian ini awalnya diperkenalkan di makalah GANSynth ICLR 2019 ( https://arxiv.org/abs/1902.08710 ).
Ukuran unduhan :
73.08 GiB
Ukuran
20.73 GiB
data :20.73 GiB
Split :
Membagi | Contoh |
---|---|
'test' | 8.518 |
'train' | 60.788 |
'valid' | 17.469 |
- Fitur :
FeaturesDict({
'audio': Audio(shape=(64000,), dtype=tf.float32),
'id': tf.string,
'instrument': FeaturesDict({
'family': ClassLabel(shape=(), dtype=tf.int64, num_classes=11),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=1006),
'source': ClassLabel(shape=(), dtype=tf.int64, num_classes=3),
}),
'pitch': ClassLabel(shape=(), dtype=tf.int64, num_classes=128),
'qualities': FeaturesDict({
'bright': tf.bool,
'dark': tf.bool,
'distortion': tf.bool,
'fast_decay': tf.bool,
'long_release': tf.bool,
'multiphonic': tf.bool,
'nonlinear_env': tf.bool,
'percussive': tf.bool,
'reverb': tf.bool,
'tempo-synced': tf.bool,
}),
'velocity': ClassLabel(shape=(), dtype=tf.int64, num_classes=128),
})
- Contoh ( tfds.as_dataframe ):
nsynth / gansynth_subset.f0_and_loudness
Deskripsi konfigurasi : Set Data NSynth terbatas pada instrumen akustik dalam interval nada MIDI [24, 84]. Menggunakan pemisahan alternatif yang memiliki instrumen yang tumpang tindih (tapi bukan catatan yang tepat) antara set kereta dan set valid / tes. Varian ini awalnya diperkenalkan di makalah GANSynth ICLR 2019 ( https://arxiv.org/abs/1902.08710 ). Versi ini juga berisi perkiraan untuk F0 menggunakan CREPE (Kim et al., 2018) dan kenyaringan persepsi berbobot A dalam desibel. Kedua sinyal disediakan pada kecepatan bingkai 250Hz.
Ukuran unduhan :
73.08 GiB
Ukuran
22.03 GiB
data :22.03 GiB
Split :
Membagi | Contoh |
---|---|
'test' | 8.518 |
'train' | 60.788 |
'valid' | 17.469 |
- Fitur :
FeaturesDict({
'audio': Audio(shape=(64000,), dtype=tf.float32),
'f0': FeaturesDict({
'confidence': Tensor(shape=(1000,), dtype=tf.float32),
'hz': Tensor(shape=(1000,), dtype=tf.float32),
'midi': Tensor(shape=(1000,), dtype=tf.float32),
}),
'id': tf.string,
'instrument': FeaturesDict({
'family': ClassLabel(shape=(), dtype=tf.int64, num_classes=11),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=1006),
'source': ClassLabel(shape=(), dtype=tf.int64, num_classes=3),
}),
'loudness': FeaturesDict({
'db': Tensor(shape=(1000,), dtype=tf.float32),
}),
'pitch': ClassLabel(shape=(), dtype=tf.int64, num_classes=128),
'qualities': FeaturesDict({
'bright': tf.bool,
'dark': tf.bool,
'distortion': tf.bool,
'fast_decay': tf.bool,
'long_release': tf.bool,
'multiphonic': tf.bool,
'nonlinear_env': tf.bool,
'percussive': tf.bool,
'reverb': tf.bool,
'tempo-synced': tf.bool,
}),
'velocity': ClassLabel(shape=(), dtype=tf.int64, num_classes=128),
})
- Contoh ( tfds.as_dataframe ):