- תיאור :
מערך הנתונים של NSynth הוא מערך שמע המכיל ~300,000 תווים מוזיקליים, כל אחד עם גובה הצליל, הגוון והמעטפה ייחודיים. כל הערה מסומנת בשלושה פיסות מידע נוספות המבוססות על שילוב של הערכה אנושית ואלגוריתמים היוריסטיים: מקור, משפחה ואיכויות.
תיעוד נוסף : חקור על ניירות עם קוד
דף הבית : https://g.co/magenta/nsynth-dataset
קוד מקור :
tfds.datasets.nsynth.Builderגרסאות :
-
2.3.0: תכונתloudness_dbחדשה בדציבלים (לא מנורמל). -
2.3.1: F0 מחושב עם תיקון נורמליזציה ב-CREPE. -
2.3.2: השתמש בתכונת אודיו. -
2.3.3(ברירת מחדל): F0 מחושב עם תיקון בנורמליזציה של גלי CREPE ( https://github.com/marl/crepe/issues/49 ).
-
שמירה אוטומטית במטמון ( תיעוד ): לא
מפתחות בפיקוח (ראה
as_superviseddoc ):Noneאיור ( tfds.show_examples ): לא נתמך.
ציטוט :
@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 (תצורת ברירת מחדל)
תיאור תצורה : ערכת הנתונים המלאה של NSynth מפוצלת לקבוצות רכבת, חוקיות ובדיקות, ללא מכשירים חופפים בין ערכת הרכבת לבין ערכות הבדיקה התקפות.
גודל הורדה :
73.07 GiBגודל מערך נתונים :
73.09 GiBפיצולים :
| לְפַצֵל | דוגמאות |
|---|---|
'test' | 4,096 |
'train' | 289,205 |
'valid' | 12,678 |
- מבנה תכונה :
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),
})
- תיעוד תכונה :
| תכונה | מעמד | צוּרָה | Dtype | תיאור |
|---|---|---|---|---|
| FeaturesDict | ||||
| שֶׁמַע | שֶׁמַע | (64000,) | לצוף32 | |
| תְעוּדַת זֶהוּת | מוֹתֵחַ | חוּט | ||
| כלי | FeaturesDict | |||
| כלי נגינה/משפחה | ClassLabel | int64 | ||
| מכשיר/תווית | ClassLabel | int64 | ||
| מכשיר/מקור | ClassLabel | int64 | ||
| גובה הצליל | ClassLabel | int64 | ||
| איכויות | FeaturesDict | |||
| תכונות/בהירות | מוֹתֵחַ | bool | ||
| תכונות/כהות | מוֹתֵחַ | bool | ||
| איכויות/עיוות | מוֹתֵחַ | bool | ||
| איכויות/דעיכה_מהירה | מוֹתֵחַ | bool | ||
| איכויות/שחרור_ארוך | מוֹתֵחַ | bool | ||
| איכויות/מולטיפוניות | מוֹתֵחַ | bool | ||
| qualities/nonlinear_env | מוֹתֵחַ | bool | ||
| איכויות/הקשה | מוֹתֵחַ | bool | ||
| איכויות/הדהוד | מוֹתֵחַ | bool | ||
| איכויות/מסונכרן טמפו | מוֹתֵחַ | bool | ||
| מְהִירוּת | ClassLabel | int64 |
- דוגמאות ( tfds.as_dataframe ):
nsynth/gansynth_subset
תיאור תצורה : מערך נתונים של NSynth מוגבל לכלים אקוסטיים במרווח הגובה של MIDI [24, 84]. משתמש בפיצולים חלופיים שיש להם חפיפה במכשירים (אך לא בהערות מדויקות) בין ערכת הרכבת לבין ערכות תקפות/בדיקה. גרסה זו הוצגה במקור במאמר ICLR 2019 GANSynth ( https://arxiv.org/abs/1902.08710 ).
גודל הורדה :
73.08 GiBגודל ערכת נתונים:
20.73 GiBפיצולים :
| לְפַצֵל | דוגמאות |
|---|---|
'test' | 8,518 |
'train' | 60,788 |
'valid' | 17,469 |
- מבנה תכונה :
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),
})
- תיעוד תכונה :
| תכונה | מעמד | צוּרָה | Dtype | תיאור |
|---|---|---|---|---|
| FeaturesDict | ||||
| שֶׁמַע | שֶׁמַע | (64000,) | לצוף32 | |
| תְעוּדַת זֶהוּת | מוֹתֵחַ | חוּט | ||
| כלי | FeaturesDict | |||
| כלי נגינה/משפחה | ClassLabel | int64 | ||
| מכשיר/תווית | ClassLabel | int64 | ||
| מכשיר/מקור | ClassLabel | int64 | ||
| גובה הצליל | ClassLabel | int64 | ||
| איכויות | FeaturesDict | |||
| תכונות/בהירות | מוֹתֵחַ | bool | ||
| תכונות/כהות | מוֹתֵחַ | bool | ||
| איכויות/עיוות | מוֹתֵחַ | bool | ||
| איכויות/דעיכה_מהירה | מוֹתֵחַ | bool | ||
| איכויות/שחרור_ארוך | מוֹתֵחַ | bool | ||
| איכויות/מולטיפוניות | מוֹתֵחַ | bool | ||
| qualities/nonlinear_env | מוֹתֵחַ | bool | ||
| איכויות/הקשה | מוֹתֵחַ | bool | ||
| איכויות/הדהוד | מוֹתֵחַ | bool | ||
| איכויות/מסונכרן טמפו | מוֹתֵחַ | bool | ||
| מְהִירוּת | ClassLabel | int64 |
- דוגמאות ( tfds.as_dataframe ):
nsynth/gansynth_subset.f0_and_loudness
תיאור תצורה : מערך נתונים של NSynth מוגבל לכלים אקוסטיים במרווח הגובה של MIDI [24, 84]. משתמש בפיצולים חלופיים שיש להם חפיפה במכשירים (אך לא בהערות מדויקות) בין ערכת הרכבת לבין ערכות תקפות/בדיקה. גרסה זו הוצגה במקור במאמר ICLR 2019 GANSynth ( https://arxiv.org/abs/1902.08710 ). גרסה זו מכילה בנוסף אומדנים עבור F0 באמצעות CREPE (Kim et al., 2018) ועוצמה תפיסתית משוקללת בדציבלים. שני האותות מסופקים בקצב פריימים של 250 הרץ.
גודל הורדה :
73.08 GiBגודל מערך נתונים :
22.03 GiBפיצולים :
| לְפַצֵל | דוגמאות |
|---|---|
'test' | 8,518 |
'train' | 60,788 |
'valid' | 17,469 |
- מבנה תכונה :
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),
})
- תיעוד תכונה :
| תכונה | מעמד | צוּרָה | Dtype | תיאור |
|---|---|---|---|---|
| FeaturesDict | ||||
| שֶׁמַע | שֶׁמַע | (64000,) | לצוף32 | |
| f0 | FeaturesDict | |||
| f0/ביטחון | מוֹתֵחַ | (1000,) | לצוף32 | |
| f0/hz | מוֹתֵחַ | (1000,) | לצוף32 | |
| f0/midi | מוֹתֵחַ | (1000,) | לצוף32 | |
| תְעוּדַת זֶהוּת | מוֹתֵחַ | חוּט | ||
| כלי | FeaturesDict | |||
| כלי נגינה/משפחה | ClassLabel | int64 | ||
| מכשיר/תווית | ClassLabel | int64 | ||
| מכשיר/מקור | ClassLabel | int64 | ||
| קוֹלָנִיוּת | FeaturesDict | |||
| עוצמת קול/db | מוֹתֵחַ | (1000,) | לצוף32 | |
| גובה הצליל | ClassLabel | int64 | ||
| איכויות | FeaturesDict | |||
| תכונות/בהירות | מוֹתֵחַ | bool | ||
| תכונות/כהות | מוֹתֵחַ | bool | ||
| איכויות/עיוות | מוֹתֵחַ | bool | ||
| איכויות/דעיכה_מהירה | מוֹתֵחַ | bool | ||
| איכויות/שחרור_ארוך | מוֹתֵחַ | bool | ||
| איכויות/מולטיפוניות | מוֹתֵחַ | bool | ||
| qualities/nonlinear_env | מוֹתֵחַ | bool | ||
| איכויות/הקשה | מוֹתֵחַ | bool | ||
| איכויות/הדהוד | מוֹתֵחַ | bool | ||
| איכויות/מסונכרן טמפו | מוֹתֵחַ | bool | ||
| מְהִירוּת | ClassLabel | int64 |
- דוגמאות ( tfds.as_dataframe ):