audiolabs / sndb

The Single Note Database, containing 30883 samples of 11 instruments from 4 datasets.

Single Note Database

The SNDB [1] contains of 30883 single tones from 4 datasets [2-5] holding 11 orchestra instruments: piano, violin, cello, contrabass, trumpet, trombone, horn, oboe, bassoon, clarinet and flute. There are 5 different dynamics (pp, p, mf, f and ff) and 42 different articulations which are explained more fully in the datasheets of the original databases [2,3,5].

Building

Please refer to build/README.md on how to create the SNDB out of the original datasets [2-5].

Usage

The file SNDB.json contains all items and their attributes:

artic	db	dyn	instr	instr_nr	is_ot	midi	multiples_nr	name	string
np	F	f_	Pia	1	False	13	0	01Pia1F013f_np___0.wav	_
np	F	m_	Pia	1	False	13	0	01Pia1F013m_np___0.wav	_
sp	R	m_	Cel	1	False	51	0	03Cel1R051m_sp___0.wav	_
tr	R	m_	Cel	1	False	56	0	03Cel1R056m_tr___0.wav	_
sp	R	f_	Cel	2	False	77	0	03Cel2R077f_sp___0.wav	_
...	...	...	...	...	...	...	...	...	...

All attributes and their meaning are described in the accompanying paper [1].

You can read this file into Pandas using

import pandas as pd

df = pd.read_json('SNDB.json')

References

1. E. F. Feichtner and B. Edler, “Description of the Single Note Database SNDB,” in Audio Engineering Society Convention 145, New York, NY, USA, 2018.
2. M. Goto, H. Hashiguchi, T. Nishimura, and R. Oka, “RWC music database: Music genre database and musical instrument sound database,” 2003.
3. F. Opolko and J. Wapnick, “Mcgill university master samples (MUMS). 11 cd-rom set,” Faculty of Music, McGill University, Montreal, Canada, 1989.
4. "Vienna Symphonic Library Special Edition 1 & 2," 2002.
5. L. Fritts, “University of Iowa musical instrument samples,” http://theremin.music.uiowa.edu/MIS.html, 1997.

sndb's Issues

Please add documentation to the repo

Dear creators of this database,
since the paper describing the database (and I assume the entries of the tables) is a bit hard to come by (...), could you kindly add some more documentation about the columns to this repository directly?

Most columns are clear from the respective (audio) databases' documentation, but some are only hard to guess, e.g. "is_ot".

(If there was a way to get the paper I'd actually do that myself)

Thanks and BR!

Where can one download the wav files?

I'd like to build this database for use in a deep learning project for educational purposes only. Are the wav files as referenced in the build/README.md available somewhere to download?

After placing all input files from all databases in a directory structure like

SNDB/
├── F/  # VSL
│   ├── 01P001Ffnp.wav
│   └── ...
├── R/  # RWC
│   ├── 011PFREP.wav
│   └── ...
├── G/  # McGill
│   ├── McGill Master Samples - CD 1 - Solo Strings an....wav
│   └── ...
├── M/  # MIS
│   ├── Bass.arco.ff.sulE.E1B1.wav
│   └── ...
└── out/