# NSynth: Neural Audio Synthesis musical notes dataset

## Facts

- Provider: Jesse Engel, Cinjon Resnick, Adam Roberts, Sander Dieleman, Douglas Eck, Karen Simonyan, Mohammad Norouzi (Google Magenta, DeepMind)
- License: CC-BY-4.0 (https://magenta.tensorflow.org/datasets/nsynth)
- Size: 40 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/jg583/NSynth
- Access: public, readable without an account
- Category: Speech & Audio
- Source: https://arxiv.org/abs/1704.01279
- Page: https://scigantic.com/archive/676a3dd0-9bba-5183-8df1-eee9e4d98a70
- Keywords: Speech & Audio, benchmark, Parquet, Hugging Face

## About

NSynth is an audio dataset of over 300,000 musical notes from 1,006 instruments from commercial sample libraries, released by Google Magenta to give audio machine learning a high-quality entry point comparable to MNIST or ImageNet. Each four-second, 16 kHz monophonic recording holds a note played for three seconds and left to decay for one second, across every pitch of a standard MIDI piano the instrument can produce at five velocities. Every note is annotated with its instrument source (acoustic, electronic or synthetic), instrument family (such as bass, brass, guitar, keyboard, mallet, organ, reed, string, vocal) and sonic qualities (bright, dark, distortion, percussive, reverb and others), from human evaluation and heuristics. It is a benchmark for neural audio synthesis (WaveNet autoencoders, GANSynth, DDSP), timbre representation learning and pitch and instrument classification. This repository is a Parquet conversion with train, validation and test folders under data/; train holds 35 Parquet batches of about 27.7 GB (measured 2026-11-14; the Hub reports about 29 GB). Load with datasets.load_dataset('jg583/NSynth') or pyarrow. Google released NSynth under CC BY 4.0; cite Engel et al., 'Neural Audio Synthesis of Musical Notes with WaveNet Autoencoders', ICML 2017.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("jg583/NSynth")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/test/test.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/676a3dd0-9bba-5183-8df1-eee9e4d98a70
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/676a3dd0-9bba-5183-8df1-eee9e4d98a70
- Over MCP: `get_schema_card` with id `676a3dd0-9bba-5183-8df1-eee9e4d98a70` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Benchmark datasets](https://scigantic.com/topics/benchmark)
- [Speech & Audio datasets](https://scigantic.com/topics/speech-and-audio)
