REF: AV DATA 2026

Audio and Video Data Licensing Partners for AI

Pangeanic is looking for data owners, production companies, audiovisual archives, research organizations and qualified collection partners able to license or produce on-camera speech data with clear rights and authorized use for commercial AI model training.

Commercial AI training rights 20,000 to 30,000 total hours Spanish up to 2,000 hours Closing date: 30 August 2026
Apply as a data partner
Project scope

Large-scale audio and video speech data for AI training

We are looking for organizations and qualified partners able to license existing rights-clean datasets or produce new audio and video speech data. Relevant sources may include conversations, interviews, meetings, podcasts, studio recordings, audiovisual archives and structured collection projects where speakers are visible and consent can be documented.

The program requires between 20,000 and 30,000 total hours. The expected language distribution is approximately 80% English and 20% across Spanish, French, Portuguese, Arabic, Hindi, Chinese, Japanese and Korean. For Spanish, the project does not require more than 2,000 hours.

This is a partnership and licensing opportunity, not a small freelance recording task. We are especially interested in data owners, rights holders, production networks and organizations able to demonstrate provenance, consent, usage rights and delivery quality at scale.

Content needed
Visible-speaker conversations: Natural speech interactions where participants appear on camera, and the audio-video relationship can be verified.
Interviews, meetings and podcasts: Recorded human speech with clear speaker turns, usage rights, stable metadata and consent documentation.
Existing archives or bespoke collection: Already licensed datasets, media libraries, research collections, controlled recordings and partner-led capture projects may all be considered.
Technical compatibility

Required formats and quality thresholds

The project requires high-quality audio and video assets prepared for model training workflows. Applicants should indicate available formats, measured quality, metadata coverage, transcript status and monthly delivery capacity.

Lossless audio WAV or FLAC, preferably 48kHz/24-bit or higher, with documentation of recording or source conditions.
Video quality Video at 720p/24fps or higher, with visible speakers and reliable lip synchronization.
Audio-video synchronization Measured AV synchronization within 40ms where technically verifiable.
Human transcripts Human transcription with segments of up to 15 seconds is preferred.
Diarization Speaker-turn information and diarization are required, or should be quoted as available add-ons.
Speaker metadata Gender and age-range metadata where legally available and properly consented.
Legal and commercial requirements

Rights-clean data with traceable consent

Commercial AI training rights

Applicants must be able to license the data for commercial AI model training and explain the scope of permitted use.

Consent and releases

Where data contains identifiable speakers, participant consent, contributor agreements or voice and image releases must be available.

Provenance and traceability

Data origin, collection method, source rights, processing steps and quality controls must be documented.

Indemnification readiness

Partners should be prepared to discuss warranties, rights representations and indemnification conditions where applicable.

This international call is subject to rights review, consent validation, applicable laws, trade sanctions and contractual feasibility.

Who should apply

We are looking for organizations with data, rights or collection capacity

  • Data owners and rights holders: organizations that have existing audio or video speech datasets available for commercial AI training.
  • Media and production companies: studios, broadcasters, podcast networks, interview producers and audiovisual archives with usable rights and quality documentation.
  • Collection partners: organizations capable of coordinating new visible-speaker recordings at scale across the required languages.
  • Transcription and data operations partners: teams that can provide transcripts, diarization, speaker metadata, and quality assurance as part of the delivery.
  • Multilingual networks: partners with reliable access to speech data in English, Spanish, French, Portuguese, Arabic, Hindi, Chinese, Japanese, and Korean.
Language coverage

Priority language distribution

The program is heavily weighted toward English, with targeted multilingual coverage for several global languages. Applicants should specify available hours, dialect or country coverage, recording conditions, consent status and delivery readiness by language.

Language Role in distribution Notes
English Approximately 80% of total volume US, UK, Ireland, Canada, Australia and other native or near-native coverage may be considered.
Spanish Up to 2,000 hours European Spanish and Latin American Spanish may be relevant, although total Spanish volume is limited.
French and Portuguese Part of the 20% multilingual allocation European and international varieties should be documented where available.
Arabic and Hindi Part of the 20% multilingual allocation Dialect, geography and recording environment should be clearly indicated.
Chinese, Japanese and Korean Part of the 20% multilingual allocation Partners should specify writing system, transcription conventions, speaker demographics and available metadata.
What to include in your application

Help us qualify your data quickly

To evaluate fit, we need concrete information about rights, quality, language coverage, metadata and delivery capacity. General descriptions are useful at the start, although final evaluation will require verifiable technical and legal detail.

01

Available volume

Total available hours by language, content type, speaker type, geography and delivery readiness.

02

Rights and consent

Commercial AI training rights, participant consent, releases, source contracts and transferability conditions.

03

Technical quality

Audio format, sample rate, bit depth, video resolution, frame rate, AV sync measurement and recording conditions.

04

Annotation layer

Transcript availability, segmentation, diarization, speaker metadata, demographic labels and QA process.

05

Monthly capacity

Number of verified hours that can be delivered per month, including ramp-up period or collection constraints.

06

Commercial terms

Indicative hourly price, minimum order, exclusivity terms, licensing limits, and delivery milestones.

Why Pangeanic

Multilingual AI data operations with legal, linguistic and technical depth

Pangeanic has worked for more than two decades at the intersection of language technology, data operations and AI deployment. Audio and video data for model training requires more than volume. It requires rights, traceability, metadata, linguistic expertise and quality control able to pass procurement review, legal review and technical ingestion.

Language data heritage

Pangeanic’s data work began with large multilingual corpora for machine translation and has evolved toward AI training data, evaluation data, alignment data and multimodal data operations.

Human-reviewed quality

The useful unit in AI data is the file plus rights, metadata, transcription, validation and a chain of decisions that can be audited.

European data governance

Pangeanic supports AI data workflows in which privacy, consent, language coverage, and deployment constraints are central to the project from the beginning.

FAQ

Frequently asked questions

Can existing datasets be submitted?
Yes. Existing datasets may be considered if the applicant can document commercial AI training rights, provenance, consent or releases where applicable, technical quality and delivery formats.
Can new on-camera speech be collected for this project?
Yes. We are also interested in qualified partners who can coordinate new visible-speaker recordings at scale, provided the process includes consent, releases, quality control, metadata, and language coverage.
Is audio-only data accepted?
The main requirement is on-camera speech with visible speakers. Audio-only corpora may be relevant for other Pangeanic data opportunities, although this specific program prioritizes audio and video speech data.
Are transcription, diarization and demographic labels required?
Human transcripts, diarization and speaker metadata are strongly preferred. If they are not already available, applicants should indicate whether they can be added, at what cost, and under which delivery timeline.
How much Spanish data is needed?
Spanish is part of the multilingual allocation, although the project does not require more than 2,000 hours of Spanish. Applicants with Spanish data should specify whether it is European Spanish, Latin American Spanish or another variety.
How are commercial terms defined?
Pricing depends on language, quality, license scope, exclusivity, metadata, transcript status, consent chain, delivery format and monthly capacity. Applicants should provide indicative per-hour pricing and minimum commercial conditions.

License or produce AV data for AI training

If your organization owns, manages, or can produce rights-clean audio and video speech data, share your available volume, language coverage, licensing rights, technical quality, and monthly delivery capacity.