- 説明:
NSynth データセットは、それぞれが固有のピッチ、音色、およびエンベロープを持つ約 300,000 の音符を含むオーディオ データセットです。各メモには、人間による評価とヒューリスティック アルゴリズムの組み合わせに基づいて、ソース、ファミリー、品質の 3 つの追加情報が注釈として付けられます。
追加のドキュメント:コードを使用したペーパーの探索
ソース コード:
tfds.datasets.nsynth.Builderバージョン:
-
2.3.0: デシベル単位の新しいloudness_db機能 (非正規化)。 -
2.3.1: CREPE の正規化修正で計算された F0。 -
2.3.2: オーディオ機能を使用します。 -
2.3.3(デフォルト): CREPE 波動正規化 ( https://github.com/marl/crepe/issues/49 ) で修正して計算された F0。
-
自動キャッシュ(ドキュメント): いいえ
監視されたキー(
as_superviseddocを参照):None図( tfds.show_examples ): サポートされていません。
引用:
@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},
}
nsynth/full (デフォルト設定)
構成の説明: 完全な NSynth データセットは、トレーニング、有効、およびテスト セットに分割され、トレーニング セットと有効/テスト セット間で機器が重複することはありません。
ダウンロードサイズ:
73.07 GiBデータセットサイズ:
73.09 GiBスプリット:
| スプリット | 例 |
|---|---|
'test' | 4,096 |
'train' | 289,205 |
'valid' | 12,678 |
- 機能構造:
FeaturesDict({
'audio': Audio(shape=(64000,), dtype=float32),
'id': string,
'instrument': FeaturesDict({
'family': ClassLabel(shape=(), dtype=int64, num_classes=11),
'label': ClassLabel(shape=(), dtype=int64, num_classes=1006),
'source': ClassLabel(shape=(), dtype=int64, num_classes=3),
}),
'pitch': ClassLabel(shape=(), dtype=int64, num_classes=128),
'qualities': FeaturesDict({
'bright': bool,
'dark': bool,
'distortion': bool,
'fast_decay': bool,
'long_release': bool,
'multiphonic': bool,
'nonlinear_env': bool,
'percussive': bool,
'reverb': bool,
'tempo-synced': bool,
}),
'velocity': ClassLabel(shape=(), dtype=int64, num_classes=128),
})
- 機能のドキュメント:
| 特徴 | クラス | 形 | Dtype | 説明 |
|---|---|---|---|---|
| 特徴辞書 | ||||
| オーディオ | オーディオ | (64000,) | float32 | |
| ID | テンソル | 弦 | ||
| 楽器 | 特徴辞書 | |||
| 楽器/ファミリー | クラスラベル | int64 | ||
| 楽器/レーベル | クラスラベル | int64 | ||
| 楽器/ソース | クラスラベル | int64 | ||
| ピッチ | クラスラベル | int64 | ||
| 資質 | 特徴辞書 | |||
| 資質/明るい | テンソル | ブール | ||
| 資質/闇 | テンソル | ブール | ||
| 品質/歪み | テンソル | ブール | ||
| 品質/fast_decay | テンソル | ブール | ||
| quality/long_release | テンソル | ブール | ||
| 品質/マルチフォニック | テンソル | ブール | ||
| quality/nonlinear_env | テンソル | ブール | ||
| 資質/パーカッシブ | テンソル | ブール | ||
| 品質/リバーブ | テンソル | ブール | ||
| 品質/テンポ同期 | テンソル | ブール | ||
| 速度 | クラスラベル | int64 |
- 例( tfds.as_dataframe ):
nsynth/gansynth_subset
構成の説明: MIDI ピッチ間隔 [24, 84] のアコースティック楽器に限定された NSynth データセット。トレーニング セットと有効/テスト セットの間でインストゥルメントが重複する (ただし正確な音符ではない) 代替分割を使用します。このバリアントは、もともと ICLR 2019 GANSynth ペーパー ( https://arxiv.org/abs/1902.08710 ) で導入されました。
ダウンロードサイズ:
73.08 GiBデータセットサイズ:
20.73 GiBスプリット:
| スプリット | 例 |
|---|---|
'test' | 8,518 |
'train' | 60,788 |
'valid' | 17,469 |
- 機能構造:
FeaturesDict({
'audio': Audio(shape=(64000,), dtype=float32),
'id': string,
'instrument': FeaturesDict({
'family': ClassLabel(shape=(), dtype=int64, num_classes=11),
'label': ClassLabel(shape=(), dtype=int64, num_classes=1006),
'source': ClassLabel(shape=(), dtype=int64, num_classes=3),
}),
'pitch': ClassLabel(shape=(), dtype=int64, num_classes=128),
'qualities': FeaturesDict({
'bright': bool,
'dark': bool,
'distortion': bool,
'fast_decay': bool,
'long_release': bool,
'multiphonic': bool,
'nonlinear_env': bool,
'percussive': bool,
'reverb': bool,
'tempo-synced': bool,
}),
'velocity': ClassLabel(shape=(), dtype=int64, num_classes=128),
})
- 機能のドキュメント:
| 特徴 | クラス | 形 | Dtype | 説明 |
|---|---|---|---|---|
| 特徴辞書 | ||||
| オーディオ | オーディオ | (64000,) | float32 | |
| ID | テンソル | 弦 | ||
| 楽器 | 特徴辞書 | |||
| 楽器/ファミリー | クラスラベル | int64 | ||
| 楽器/レーベル | クラスラベル | int64 | ||
| 楽器/ソース | クラスラベル | int64 | ||
| ピッチ | クラスラベル | int64 | ||
| 資質 | 特徴辞書 | |||
| 資質/明るい | テンソル | ブール | ||
| 資質/闇 | テンソル | ブール | ||
| 品質/歪み | テンソル | ブール | ||
| 品質/fast_decay | テンソル | ブール | ||
| quality/long_release | テンソル | ブール | ||
| 品質/マルチフォニック | テンソル | ブール | ||
| quality/nonlinear_env | テンソル | ブール | ||
| 資質/パーカッシブ | テンソル | ブール | ||
| 品質/リバーブ | テンソル | ブール | ||
| 品質/テンポ同期 | テンソル | ブール | ||
| 速度 | クラスラベル | int64 |
- 例( tfds.as_dataframe ):
nsynth/gansynth_subset.f0_and_loudness
構成の説明: MIDI ピッチ間隔 [24, 84] のアコースティック楽器に限定された NSynth データセット。トレーニング セットと有効/テスト セットの間でインストゥルメントが重複する (ただし正確な音符ではない) 代替分割を使用します。このバリアントは、もともと ICLR 2019 GANSynth ペーパー ( https://arxiv.org/abs/1902.08710 ) で導入されました。このバージョンには、CREPE (Kim et al., 2018) を使用した F0 の推定値と、デシベル単位の A 加重知覚ラウドネスも含まれています。両方の信号は、250Hz のフレーム レートで提供されます。
ダウンロードサイズ:
73.08 GiBデータセットサイズ:
22.03 GiBスプリット:
| スプリット | 例 |
|---|---|
'test' | 8,518 |
'train' | 60,788 |
'valid' | 17,469 |
- 機能構造:
FeaturesDict({
'audio': Audio(shape=(64000,), dtype=float32),
'f0': FeaturesDict({
'confidence': Tensor(shape=(1000,), dtype=float32),
'hz': Tensor(shape=(1000,), dtype=float32),
'midi': Tensor(shape=(1000,), dtype=float32),
}),
'id': string,
'instrument': FeaturesDict({
'family': ClassLabel(shape=(), dtype=int64, num_classes=11),
'label': ClassLabel(shape=(), dtype=int64, num_classes=1006),
'source': ClassLabel(shape=(), dtype=int64, num_classes=3),
}),
'loudness': FeaturesDict({
'db': Tensor(shape=(1000,), dtype=float32),
}),
'pitch': ClassLabel(shape=(), dtype=int64, num_classes=128),
'qualities': FeaturesDict({
'bright': bool,
'dark': bool,
'distortion': bool,
'fast_decay': bool,
'long_release': bool,
'multiphonic': bool,
'nonlinear_env': bool,
'percussive': bool,
'reverb': bool,
'tempo-synced': bool,
}),
'velocity': ClassLabel(shape=(), dtype=int64, num_classes=128),
})
- 機能のドキュメント:
| 特徴 | クラス | 形 | Dtype | 説明 |
|---|---|---|---|---|
| 特徴辞書 | ||||
| オーディオ | オーディオ | (64000,) | float32 | |
| f0 | 特徴辞書 | |||
| f0/信頼度 | テンソル | (1000,) | float32 | |
| f0/hz | テンソル | (1000,) | float32 | |
| f0/ミディ | テンソル | (1000,) | float32 | |
| ID | テンソル | 弦 | ||
| 楽器 | 特徴辞書 | |||
| 楽器/ファミリー | クラスラベル | int64 | ||
| 楽器/レーベル | クラスラベル | int64 | ||
| 楽器/ソース | クラスラベル | int64 | ||
| ラウドネス | 特徴辞書 | |||
| ラウドネス/デシベル | テンソル | (1000,) | float32 | |
| ピッチ | クラスラベル | int64 | ||
| 資質 | 特徴辞書 | |||
| 資質/明るい | テンソル | ブール | ||
| 資質/闇 | テンソル | ブール | ||
| 品質/歪み | テンソル | ブール | ||
| 品質/fast_decay | テンソル | ブール | ||
| quality/long_release | テンソル | ブール | ||
| 品質/マルチフォニック | テンソル | ブール | ||
| quality/nonlinear_env | テンソル | ブール | ||
| 資質/パーカッシブ | テンソル | ブール | ||
| 品質/リバーブ | テンソル | ブール | ||
| 品質/テンポ同期 | テンソル | ブール | ||
| 速度 | クラスラベル | int64 |
- 例( tfds.as_dataframe ):