Datasets:
license: other
license_name: trf-evaluation-license
license_link: LICENSE.md
pretty_name: The Rights Foundry Collection One
task_categories:
- audio-classification
tags:
- audio
- music
- music-dataset
- music-information-retrieval
- music-analysis
- audio-classification
- machine-learning
- generative-ai
- music-ai
- ai-training
- ai-evaluation
- benchmarking
- licensed-data
- rights-cleared
- commercial-music
gated: manual
extra_gated_heading: >-
The License must be accepted to allow access to this Dataset. NOTE we are
currently not accepting external requests, pending final validation of the
dataset
extra_gated_prompt: >-
### License Agreement
**THE RIGHTS FOUNDRY COLLECTION ONE**
Non-Commercial Research and Evaluation License
**1\. Acceptance.** By requesting access to The Rights Foundry Collection One
(the Corpus, meaning the audio recordings, metadata, captions, and embedded
identifiers together), you agree to these terms with The Rights Foundry, Inc.,
a Delaware corporation (TRF). If you are accepting on behalf of an
organization, you confirm that you have authority to bind it, and "you" means
that organization.
**2\. Definitions.** "Non-Commercial Use" means academic research, education,
personal study, analysis, benchmarking, model evaluation, and the training of
experimental models that are not deployed, distributed, or made available
commercially in any form. Internal research or testing conducted by or for a
commercial entity to evaluate, develop, improve, or support a current or
planned product, service, model, or revenue generating activity is Commercial
Use, except that a commercial entity may conduct limited internal technical
evaluation solely to determine whether to obtain a commercial license from TRF
("Evaluation Use"). Under Evaluation Use, a commercial entity may create
temporary test models for that evaluation, but may not deploy or distribute
them, use them or their results to develop or improve a product or service, or
retain reusable model weights after the evaluation unless TRF grants a
commercial license. "Commercial Use" means any other use of the Corpus, in
whole or in part, intended for or directed toward commercial advantage or
monetary compensation, by you or by anyone you enable.
**3\. License grant.** TRF grants you a limited, non-exclusive,
non-transferable, non-sublicensable, worldwide, royalty-free license,
revocable as provided in Section 15, to download and use the Corpus solely for
Non-Commercial Use and Evaluation Use.
**4\. Permitted research activity.** You may analyze the Corpus, extract
features from it, benchmark and evaluate models against it, and train
experimental models on it. You may publish papers, benchmark results,
aggregate statistics, research code, documentation, and non-commercial
research demonstrations based on the Corpus, provided no demonstration gives
third parties systematic access to generate, retrieve, or download audio. You
may distribute model weights and non-reconstructive feature representations
created through permitted research only after reasonable testing confirms that
they do not contain, reproduce, expose, or permit reconstruction of any
recording or any substantial portion of one, and that they cannot be used to
replicate, imitate, or simulate any voice or distinctive sound described in
Section 7\. They must be accompanied by the required Corpus attribution and
distributed under this license or terms at least as protective of Sections 4
through 7, 11, and 12\.
**5\. Prohibited uses.** No Commercial Use of any kind, of the Corpus or of
anything derived from it, in whole or in part. Without limiting that, you will
not: train or fine tune any model that is or will be made commercially
available or deployed in a product or service, including through a hosted or
software as a service inference endpoint or API; publish, release, or monetize
any output that reproduces, is recognizably based on, or permits
reconstruction of a Corpus recording or a substantial portion of one;
redistribute, re-host, mirror, or include the Corpus, any track, or any
portion of the metadata or captions in another dataset or collection;
distribute any track to any streaming, download, or user generated content
platform; sell, rent, or transfer the Corpus or any track; or create
derivative musical works, remixes, or edits for release. Section 7 sets
further prohibitions on replicating voices and distinctive sounds, which apply
in addition to this Section. You will comply with applicable law and respect
third party rights in all use of the Corpus.
**6\. Rights systems.** You will not register or attempt to register any
Corpus recording, excerpt, fingerprint, derivative, model output, or
associated rights information with any content identification system, digital
distributor, performing rights organization, neighboring rights organization,
copyright registry, or rights management service, and you will not assert
ownership of, or initiate any claim, takedown, or monetization action relating
to, any Corpus material.
**7\. Voice, sound, and identity.** The Corpus contains the recorded voices,
performances, and signature sounds of artists and musicians. You may not use
the Corpus or anything derived from it to train, fine tune, condition, or
prompt any model, system, or output that replicates, imitates, clones, or
simulates the voice, singing voice, vocal timbre, instrumental performance, or
other distinctive sound of any artist, performer, or recording in the Corpus,
whether or not that person or sound is named or identifiable to you. This
includes voice cloning, voice conversion, speaker or singer embeddings used
for generation, timbre transfer, and instrument or sound models derived from
Corpus recordings. Nothing in this license lifts this restriction. Any
commercial use requires a separate written license from TRF and any performer
consent or other rights required by applicable law, and a commercial license
from TRF does not by itself grant rights in any performer’s voice or likeness.
Non-Commercial Use under Section 4 remains available for analysis,
recognition, classification, and benchmarking that does not create a
generative capability described in this Section. You may not use the Corpus or
anything derived from it to impersonate, deceive, harass, exploit, or falsely
suggest the participation, approval, or endorsement of any person. You may not
deploy the Corpus or anything derived from it for biometric identification,
authentication, surveillance, eligibility, employment, credit, insurance,
healthcare, or other decisions concerning an identifiable person. This license
does not authorize commercial voice cloning or any other Commercial Use.
**8\. Ownership and licensing authority.** Each master recording and musical
composition in the Corpus is owned outright by a single source rights holder.
TRF has the rights and authority necessary to make the Corpus available and to
grant this license, and TRF is the licensor under this license. No ownership
in any recording, composition, performance, or metadata transfers to you, and
all rights not expressly granted are reserved. This license grants no right in
the voice, name, or likeness of any performer.
**9\. Your research materials.** As between you and TRF, you retain ownership
of your independently developed research code, analyses, model architectures,
model weights, and research results. That ownership does not extend to the
Corpus or any recording, composition, or metadata, and does not remove the
restrictions of this license.
**10\. Organization access.** If you accept for an organization, the
organization may permit its employees, students, researchers, and individual
contractors to use the Corpus solely for the organization’s permitted use
under this license, provided they are informed of and bound by these terms.
The organization is responsible for their compliance. Access credentials may
not be shared.
**11\. Integrity and audit.** You will not remove, alter, or obscure any
metadata, rights information, or embedded identifier in the files, and you
will keep the rights information associated with any excerpt or feature set
you create from the Corpus. On TRF’s written request, not more than once in
any twelve month period, you will confirm in writing that you are in
compliance with Sections 4 through 7\. You acknowledge that the Corpus may
contain unique identifiers and reference recordings that TRF uses to detect
unauthorized use. If TRF reasonably believes you have breached Section 5, 6,
or 7, you will cooperate in good faith with a reasonable review of your
compliance, including by describing in reasonable detail any model you have
trained on or evaluated with the Corpus and the testing performed under
Section 4, and by making relevant model documentation and training records
available. TRF will keep any model documentation and training records you
provide confidential and use them only to assess compliance and enforce this
license.
**12\. Attribution.** Publications, public benchmarks, and model documentation
that use the Corpus must cite it as described on the dataset card.
**13\. Commercial licensing.** This license grants no Commercial Use rights.
Commercial Use requires a separate written commercial license from TRF,
together with any performer consent or other rights required by applicable
law. To request a commercial license, including for model training intended
for production deployment, visit
[https\://www\.therightsfoundry.com/huggingfacecollectionone](https://www.therightsfoundry.com/huggingfacecollectionone)
or contact
[licensing@therightsfoundry.com](mailto:licensing@therightsfoundry.com).
**14\. Access information.** You acknowledge that the access information you
provide through the platform gate (such as your name, email address, and
organization) is shared with TRF, which may use it to administer this license
and to contact you about licensing, in accordance with TRF’s privacy policy at
therightsfoundry.com/privacy.
**15\. Term and termination.** This license continues until terminated. TRF
may terminate immediately upon unauthorized redistribution, Commercial Use, or
breach of Section 6 or Section 7\. For any other remediable breach, TRF will
give notice and thirty days to cure before termination. TRF may withdraw
individual recordings from the Corpus where legally required or where a rights
claim arises, and you will delete withdrawn recordings on notice. On
termination you will delete all copies of the Corpus in your possession or
control. Papers, benchmark results, and aggregate statistics properly
published during the term may remain available, and model weights properly
created and distributed under Section 4 during the term may continue to be
used and distributed on the same non-commercial terms. "Noncompliant
Materials" means any model, model weights, checkpoint, speaker or singer
embedding, feature set, output, or other material created, used, or
distributed in breach of Sections 4 through 7, including anything prohibited
by Section 7\. Section 9 does not protect Noncompliant Materials. Upon
termination, or upon your discovery or notice of a violation, whichever is
earlier, and notwithstanding the cure period above, you will: (a) immediately
cease all use and distribution of Noncompliant Materials; (b) within \[ten\]
business days, delete or permanently disable all Noncompliant Materials in
your possession or control and remove every copy you have made available to
others, including from public repositories; (c) within \[ten\] business days,
notify each downstream recipient that you know of or can reasonably identify
of the violation and instruct them to cease use and distribution and to delete
or permanently disable the affected materials; and (d) within \[fifteen\]
business days, certify your compliance with this paragraph to TRF in writing.
You may preserve evidence of the violation only where the law requires, with
access restricted to those who need it for that purpose, no further use or
distribution except as the law requires, and deletion when the preservation
obligation ends. These obligations survive termination and apply only to
Noncompliant Materials; they do not change the treatment of compliant research
materials described above. Sections 5 through 9, 11 through 17, and any other
provisions that by their nature should survive termination, survive
termination.
**16\. Disclaimers and limitation of liability.** Except for TRF’s express
statements in Section 8 regarding ownership of the recordings and compositions
and its authority to grant this license, the Corpus is provided as is, without
warranties of any kind, express or implied, including title, non-infringement,
merchantability, and fitness for a particular purpose. To the maximum extent
permitted by law, neither TRF nor its licensors, catalog partners, or rights
holders will be liable for indirect, consequential, incidental, special,
exemplary, or punitive damages, or for lost profits, data, goodwill, or
business opportunity, and TRF’s total aggregate liability arising out of this
license is limited to one hundred US dollars. TRF’s licensors, catalog
partners, and rights holders are intended beneficiaries of this Section.
**17\. General.** This license is governed by the laws of the State of
California, excluding its conflict of law rules, and the state and federal
courts located in Santa Clara County, California have exclusive jurisdiction
over any dispute arising from it. If any provision is unenforceable, the
remainder stands. No failure to enforce is a waiver. You may not assign this
license. This license is the entire agreement between you and TRF concerning
the Corpus.
### By proceeding, you accept these license conditions.
extra_gated_description: >-
Please provide your affiliation and desired use case. Review may take 48
hours.
extra_gated_fields:
Company/University: text
Intended Use Case:
type: select
options:
- Academic Research
- Evaluation for Commercial Product use
- Internal Testing
I agree to use this dataset for non-commercial purposes only: checkbox
I agree not to redistribute the audio or metadata: checkbox
I agree not to register any of this content in rights databases: checkbox
I agree not to use this dataset for voice or sound cloning: checkbox
extra_gated_button_content: I Agree to the License Terms
dataset_info:
features:
- name: audio
dtype: audio
- name: track_id
dtype: string
- name: artist_id
dtype: string
- name: feat_artist_ids
dtype: string
- name: duration
dtype: float64
- name: bpm
dtype: int64
- name: key
dtype: string
- name: genres
dtype: string
- name: subGenres
dtype: string
- name: moods
dtype: string
- name: instrumentation
dtype: string
- name: vocal_type
dtype: string
- name: explicit
dtype: bool
- name: use_case
dtype: string
- name: release_year
dtype: string
- name: keywords
dtype: string
- name: caption
dtype: string
- name: rights_line
dtype: string
splits:
- name: train
num_bytes: 11389140371
num_examples: 500
download_size: 11390152424
dataset_size: 11389140371
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
The Rights Foundry Collection One
Dataset Summary
A licensed 500-track music dataset for non-commercial AI research, reproducible benchmarking, model evaluation, and music-information-retrieval research.
The Rights Foundry Collection One contains 500 fully rights-cleared commercial music tracks from independent artists, provided as 16-bit, 44.1 kHz stereo WAV audio with structured descriptive metadata.
The dataset is designed for non-commercial academic research, reproducible benchmarking, non-commercial evaluation, and peer-reviewed model evaluations under the custom, restrictive Non-Commercial Research and Evaluation License.
Need a larger licensed music corpus?
The Rights Foundry's catalog contains 100,000+ commercially released tracks, including 35,000+ tracks cleared for AI use under signed Generative AI Addenda.
These larger datasets support licensing discussions for AI training, fine-tuning, evaluation, retrieval, generative applications, and downstream commercial use, subject to the applicable rights and agreements.
Explore The Rights Foundry
Need more than 500 tracks? Contact us: hello@therightsfoundry.com
Dataset at a Glance
| Parameter | Value |
|---|---|
| Tracks: | 500 |
| Total duration: | 17.9 hours |
| Licence: | Custom Non-Commercial Research and Evaluation License (LICENSE.md) |
| Territory: | Global |
| Audio quality: | 16-bit, 44.1 kHz, 2-channel stereo |
| Format: | Parquet, with WAV audio embedded alongside the metadata |
| Vocals: | About half the tracks are instrumental; the rest have male, female, background, choral, or duet vocals |
| Primary use: | Non-commercial academic research and evaluation |
| Access: | Request Access on Hugging Face (requests are reviewed manually) |
What is this dataset?
The Rights Foundry Collection One is a licensed music dataset for AI and machine-learning research.
It contains 500 fully rights-cleared commercial music tracks, together with structured metadata describing musical and audio characteristics including genre, sub-genre, mood, instrumentation, vocals, BPM, musical key, duration, explicit content, keywords, and textual captions.
The corpus is intended to provide researchers with a defined, reproducible collection of commercially released music for audio research, music-information retrieval, benchmarking, model evaluation, and related academic research.
All use of the dataset is subject to the Non-Commercial Research and Evaluation License, which must be agreed to by clicking Request Access on the Hugging Face dataset repository and accepting the terms.
What can this dataset be used for?
The dataset is licensed for specified non-commercial uses, including:
- Non-commercial academic research, education, and personal study
- Reproducible benchmarking
- Non-commercial model evaluation
- Peer-reviewed model evaluations
- Training experimental models that are not deployed, distributed, or made available commercially, and that cannot replicate any voice or distinctive sound in the dataset
- Music-information retrieval research
- Music and audio analysis research
- Limited internal technical evaluation by a commercial entity, solely to decide whether to obtain a commercial license from The Rights Foundry
Researchers must comply with the full license when using the dataset.
Can this dataset be used for commercial AI training?
No.
The 500-track Rights Foundry Collection One is provided under a custom, restrictive non-commercial license. It is not licensed for commercial deployment, commercial exploitation, SaaS inference integration, or model-weight monetisation.
For commercial AI training, fine-tuning, evaluation, retrieval, generative applications, or other commercial use cases, a separate written commercial license from The Rights Foundry is required.
Need more than 500 tracks or expanded usage rights? Visit therightsfoundry.com/huggingfacecollectionone or contact: licensing@therightsfoundry.com
Restrictions
The license prohibits:
- Commercial use of any kind, of the dataset or anything derived from it
- Training or fine-tuning any model that is or will be commercially available or deployed, including through a hosted or SaaS inference endpoint or API
- Publishing, releasing, or monetising outputs that reproduce, are recognisably based on, or permit reconstruction of a recording
- Redistributing, re-hosting, or mirroring the audio, metadata, or captions, or including them in another dataset
- Distributing any track to streaming, download, or user-generated-content platforms
- Creating derivative musical works, remixes, or edits for release
- Replicating, imitating, cloning, or simulating any voice or distinctive sound in the dataset (see below)
- Registering any recording, excerpt, fingerprint, derivative, or model output with a content identification system, distributor, or rights organisation
- Removing or altering metadata, rights information, or embedded identifiers
Voice, sound, and biometric restrictions
The dataset contains the recorded voices, performances, and signature sounds of artists and musicians. The license prohibits using the dataset, or anything derived from it, to train, fine-tune, condition, or prompt any model, system, or output that replicates, imitates, clones, or simulates the voice, singing voice, vocal timbre, instrumental performance, or other distinctive sound of any artist, performer, or recording in the dataset. This includes:
- Voice cloning and voice conversion
- Speaker or singer embeddings used for generation
- Timbre transfer
- Instrument or sound models derived from the recordings
The license also prohibits:
- Impersonating, deceiving, harassing, or exploiting any person, or falsely suggesting their participation or endorsement
- Biometric identification, authentication, surveillance, or decisions concerning an identifiable person
- Commercial voice cloning
Non-commercial analysis, recognition, classification, and benchmarking remain permitted, provided they do not create any of the generative capabilities above.
Users should review the complete license before using any track for model training, fine-tuning, evaluation, or other computational purposes.
Dataset Structure
The dataset is distributed as a single train split in Parquet format, divided into 23 files. Each row is one track: the audio column holds the original WAV file, and the remaining columns hold its metadata.
TRF-Collection-One/
├── README.md
├── LICENSE.md
└── data/
├── train-00000-of-00023.parquet
├── train-00001-of-00023.parquet
└── ...
Once your access request has been approved, log in with hf auth login and load the dataset with:
from datasets import load_dataset
ds = load_dataset("The-Rights-Foundry/TRF-Collection-One", split="train")
The full download is about 11 GB. To read tracks without downloading everything first, pass streaming=True.
Audio Format
The audio is provided as:
- WAV
- 16-bit
- 44.1 kHz sample rate
- 2-channel stereo
Metadata Fields
Each row contains the following fields
Audio:
audio(audio): The track's WAV audio.
Descriptive Metadata:
track_id(string): Unique track identifier (example:001)artist_id(string): Identifier of the track's primary artist (example:1)feat_artist_ids(string|null): Comma-separated identifiers of any featured artists (example:"14, 5"). Blank where there are none.duration(number): Duration in seconds. Fractional on almost every track.bpm(integer): Editorial beats per minute. (example:115)key(string): Primary musical key (example:"D minor")genres(string): Comma-separated list of genres (example:"hip hop, pop")subGenres(string): Comma-separated list of sub-genres (example:"dance pop, electropop")moods(string): Comma-separated list of moods (example:"exciting, happy, swagger")instrumentation(string): Comma-separated list of main instruments (partial example:"brass, claps stomps, electric guitars, electronic")vocal_type(string): Editorial vocal type, lowercase. One ofinstrumental,male,female,background,choral,duet.explicit(boolean): Explicit Language/Themes flag. No explicit content is included in this dataset (example:false)use_case(string): Suggested usage area. One ofCinematic,Creator Tools,Fitness & wellness,Gaming,Retail & brand.release_year(string|null): Release year, if formally released (example:"2022")keywords(string): Comma-separated list of keywords describing this track (partial example:"pop, electropop, positive")caption(string): Textual description of the track (partial example:"Pulsing synth lines anchor this instrumental cue, its driving electronic current deepened by a steady bass pulse")rights_line(string): Rights under which this track is included (example:"Master and composition owned and controlled by Audiosocket. Rights granted by The Rights Foundry, as sublicensor, under the dataset license.")
Frequently Asked Questions
How many tracks are in The Rights Foundry Collection One?
The Rights Foundry Collection One contains 500 fully rights-cleared commercial music tracks.
Is The Rights Foundry Collection One free to use?
The dataset is available for specified non-commercial academic research, reproducible benchmarking, non-commercial evaluation, and peer-reviewed model evaluations, subject to the custom Non-Commercial Research and Evaluation License.
Is this a music dataset for AI research?
Yes. The collection is a licensed commercial music dataset for AI and machine-learning research, including research involving audio analysis, music-information retrieval, benchmarking, and model evaluation, subject to the license.
Can I use the dataset for commercial AI training?
No. The 500-track collection is subject to a restrictive non-commercial license. Contact The Rights Foundry for separately licensed commercial AI datasets and licensing options: hello@therightsfoundry.com
Does the dataset contain commercial music?
Yes. The collection contains 500 fully rights-cleared commercial music tracks.
What audio format is provided?
The tracks are provided as 16-bit, 44.1 kHz, 2-channel stereo WAV audio.
How can I access the dataset?
Click Request Access on the Hugging Face repository and accept the terms of the custom Non-Commercial Research and Evaluation License.
Where can I find a larger licensed music dataset?
- The Rights Foundry's production catalog contains 100,000+ commercially released tracks, including 35,000+ tracks cleared for AI use under signed Generative AI Addenda.
- Explore The Rights Foundry → www.therightsfoundry.com
- For larger datasets and licensing enquiries, contact: hello@therightsfoundry.com
Need More Than 500 Tracks?
The Rights Foundry Collection One is the 500-track research and evaluation collection from The Rights Foundry.
For projects requiring a larger licensed music corpus, The Rights Foundry's production catalog contains:
- 100,000+ commercially released tracks
- 35,000+ tracks cleared for AI use under signed Generative AI Addenda
- 3,500+ independent artists
The AI-cleared catalog is available for licensing discussions covering use cases such as AI training, fine-tuning, evaluation, retrieval, generative applications, and downstream commercial rights, subject to the applicable agreements.
For larger dataset and licensing enquiries: hello@therightsfoundry.com
About The Rights Foundry
The Rights Foundry provides music rights infrastructure for digital products and AI pipelines.
The Rights Foundry provides an API layer that clears, enforces, meters, and monetises music rights at the point of use, for digital products and AI pipelines.
- Category: Music rights infrastructure
- Founded: 2026
- Catalog: 100,000+ tracks
- AI-cleared: 35,000+ tracks
- Artists: 3,500+ independent artists
Citation
Publications, public benchmarks, and model documentation that use this dataset must cite it as follows (Section 12 of the license):
The Rights Foundry, Inc. (2026). The Rights Foundry Collection One [Dataset]. Hugging Face. https://huggingface.co/datasets/The-Rights-Foundry/TRF-Collection-One
@misc{therightsfoundry2026collectionone,
author = {{The Rights Foundry, Inc.}},
title = {The Rights Foundry Collection One},
year = {2026},
publisher = {Hugging Face},
howpublished = {\url{https://huggingface.co/datasets/The-Rights-Foundry/TRF-Collection-One}},
note = {Licensed under the Non-Commercial Research and Evaluation License}
}