PREMIUM AUDIO DATA

Voice data for the languages everyone else skips.

40,000 vetted freelancers across every EU language and the dialects that matter — Jutlandic Danish, Swiss German, Galician, Sámi, Frisian, and 90+ more. Audit-ready for the EU AI Act, GDPR-compliant by design.

See coverage matrix

LIVE DEMO

“Jeg DANISH:jutlandic købte tre hundrede kroner CURRENCY værd i Århus.” LOCATION

Confidence: 99.2% QA: 2 reviewers

The bottleneck isn’t model capability. It’s training data your speakers actually speak.

24 ¹ EU official
languages
90+ ² regional dialects
+ minority langs
40,000 ³ vetted
freelancers

Norwegian Nynorsk

1,420 vetted

Western dialects, Inland

Details

Concentrated in Vestlandet + Inland counties. ~38% female / 62% male active, ages 22–58. Common utterance length: 4–18s. Used to fix ASR bias against non-Bokmål Norwegian.

Jutlandic Danish

740 vetted

Western, Sønderjysk

Details

Sønderjysk (South Jutlandic) has high regional variance. Speakers vetted across Aalborg, Aarhus, Esbjerg, and Tønder regions. Often missing from off-the-shelf Danish ASR.

Swiss German

985 vetted

Zürich, Bern, Basel variants

Details

All three major Alemannic urban variants plus rural fallbacks. High demand from finance + insurance NLU teams. Most utterance prompts span finance, healthcare, and consumer domains.

Galician

520 vetted

Northern + Coastal

Details

Co-official with Spanish in Galicia. Coastal vs interior phonology differs meaningfully. Used by clients building bilingual ES/GL voice products.

Sámi

210 vetted

Northern, Inari, Skolt

Details

Three of the nine Sámi languages. Northern is most populous; Inari + Skolt are critically endangered. Recruited via community liaison networks across Finnmark + Inari + Sevettijärvi.

Frisian

285 vetted

West, North, Saterland

Details

West Frisian (Netherlands), North Frisian (Schleswig-Holstein), Saterland Frisian (Lower Saxony). Used in regional government accessibility + cultural-heritage projects.

Welsh

640 vetted

Northern + Southern

Details

North Welsh + South Welsh — the dialect boundary is meaningful for ASR. Common downstream uses: government services, BBC Cymru content tooling, education.

Corsican

155 vetted

Northern + Southern

Details

Cismontano (north) and Oltramontano (south). Italo-Romance lineage. Among the smaller pools — typically scoped 3–6 weeks ahead for collection.

See all 24 EU languages + 90+ dialects 59 entries

Language	Region / Country	Vetted speakers	Dialects covered	Sample audio
Germanic 13
English	IE, MT (EU official)	2,900	Hiberno-English, Maltese English	scoping call
German	DE, AT, BE-DG, LU	2,500	Standard, Austrian, Low German, Swiss German (Zürich/Bern/Basel)	scoping call
Dutch	NL, BE-VL	1,300	Hollands, Flemish, Brabantian, Limburgish	scoping call
Danish	DK	1,000	Standard, Jutlandic (Western), Sønderjysk, Bornholmian	scoping call
Swedish	SE, FI	1,200	Standard, Skånsk, Gotlandic, Finland-Swedish	scoping call
Finnish	FI	950	Standard, Eastern, Western, Helsinki slang	scoping call
Norwegian (Bokmål)	NO (non-EU but EEA)	1,850	Eastern, Northern	scoping call
Norwegian (Nynorsk)	NO (non-EU but EEA)	1,420	Western, Inland	scoping call
Faroese	FO	145	Tórshavn, Suðuroy	scoping call
Frisian (West)	NL	215	Wood Frisian, Clay Frisian	scoping call
Frisian (North)	DE	75	Mooring, Fering	scoping call
Frisian (Saterland)	DE	55	Seelter	scoping call
Yiddish	EU-wide	105	Litvish, Galitzish	scoping call
Romance 15
French	FR, BE, LU	2,800	Métropolitain, Belgian, Walloon, Acadian	scoping call
Italian	IT, MT	2,350	Standard, Sicilian, Neapolitan, Venetian, Sardinian, Friulian	scoping call
Spanish (Castilian)	ES	2,700	Castilian, Andalusian, Murcian, Canarian	scoping call
Catalan	ES, AD	850	Central, Valencian, Balearic	scoping call
Galician	ES	520	Northern, Coastal	scoping call
Portuguese	PT	1,150	European Portuguese, Azorean, Madeiran	scoping call
Romanian	RO	1,100	Standard, Moldavian, Transylvanian	scoping call
Asturian	ES	220	Central, Western, Eastern	scoping call
Sardinian	IT	235	Logudorese, Campidanese	scoping call
Corsican	FR	155	Cismontano, Oltramontano	scoping call
Sicilian	IT	410	Palermitan, Catanese	scoping call
Friulian	IT	155	Central, Western, Carnico	scoping call
Romansh	CH (non-EU)	115	Sursilvan, Vallader, Puter, Rumantsch Grischun	scoping call
Occitan	FR, IT, ES	230	Gascon, Languedocien, Provençal	scoping call
Walloon	BE	105	Central, Eastern	scoping call
Celtic 6
Irish Gaelic	IE (EU official)	410	Connacht, Munster, Ulster	scoping call
Welsh	UK (non-EU)	640	Northern, Southern	scoping call
Scottish Gaelic	UK (non-EU)	220	Hebridean, Highland	scoping call
Manx	IM (non-EU)	55	Revival cohort	scoping call
Breton	FR	225	Kerneveg, Leoneg, Tregerieg, Gwenedeg	scoping call
Cornish	UK (non-EU)	55	Kernewek Kemmyn revival	scoping call
Slavic 11
Polish	PL	1,950	Standard, Silesian, Kashubian-adjacent	scoping call
Czech	CZ	1,150	Standard, Moravian	scoping call
Slovak	SK	850	Standard, Eastern, Central	scoping call
Slovenian	SI	620	Standard, Prekmurje	scoping call
Bulgarian	BG	1,100	Eastern, Western	scoping call
Croatian	HR	1,100	Standard (Štokavian), Kajkavian, Chakavian	scoping call
Kashubian	PL	145	Northern, Southern	scoping call
Silesian	PL	220	Upper Silesian	scoping call
Sorbian (Upper)	DE	95	Bautzen / Budyšin	scoping call
Sorbian (Lower)	DE	55	Cottbus / Chóśebuz	scoping call
Rusyn	SK, PL, HU	95	Carpatho-Rusyn	scoping call
Uralic 9
Hungarian	HU	1,450	Standard, Palóc, Csángó	scoping call
Estonian	EE	850	Standard, Võro, South Estonian	scoping call
Finnish (Karelian-adjacent)	FI	— see Finnish row above	see Germanic-Finnish overlap	scoping call
Sámi (Northern)	NO, SE, FI	145	Davvisámegiella	scoping call
Sámi (Lule)	NO, SE	55	Julevsámegiella	scoping call
Sámi (Inari)	FI	40	Anarâškielâ	scoping call
Sámi (Skolt)	FI, RU border cohort	40	Nuõrttsääʹmǩiõll	scoping call
Karelian	FI, RU border cohort	55	Livvi-Karelian, Northern Karelian	scoping call
Võro	EE	95	South Estonian	scoping call
Baltic 2
Latvian	LV	620	Standard, Latgalian	scoping call
Lithuanian	LT	620	Standard, Žemaitian, Aukštaitian	scoping call
Hellenic 1
Greek	GR, CY	1,100	Standard, Cypriot Greek	scoping call
Other / Isolates 2
Maltese	MT	410	Standard, regional	scoping call
Basque (Euskara)	ES, FR	410	Biscayan, Gipuzkoan, Upper Navarrese, Lapurdian	scoping call

Counts refreshed quarterly. Last refresh: 2026-Q1. Live capacity may vary by 5-10% based on freelancer availability per project window. Vetted = ID-verified, audio-quality-tested, NDA-signed.

Audio samples: native-speaker recordings from Wikimedia Lingua Libre and Wikitongues, licensed CC-BY-SA 4.0. Corsican sample is TTS-generated pending real recording.

01 Input

Brief & spec

You arrive with a use case. We arrive with a list of languages, dialects, accents, demographic distribution, recording conditions, and acceptance criteria. We converge in 1–2 calls.

24 EU languages
90+ dialects
Speaker demographics
Acoustic conditions
Acceptance criteria

02 Process

Pipeline, end to end

01 Brief spec specs locked languages, dialects, demographics, acceptance criteria
02 Freelancer matching pool of 40k filtered to qualified subset native-speaker validation, dialect coverage, region
03 Record studio-grade or remote-with-spec; quality validated per file 44.1 kHz / 16-bit min, SNR + clipping checks
04 Annotation orthographic + phonetic + speaker meta + region IPA transcription, speaker ID, accent label
05 QA + Audit Cohen’s Kappa ≥0.85 inter-annotator agreement EU AI Act Article 10 doc auto-generated

¹ 0.85 is the industry-standard threshold for premium audio data. Measured continuously, reported in the QA dashboard. Methodology shared in the scoping call.

manifest.json | json

{
  "dataset_id": "ypai-NO-nyn-2026Q2",
  "language": "nob",
  "dialect": "nynorsk-western",
  "speakers": 142,
  "hours": 218.7,
  "kappa": 0.87,
  "wer_baseline": 9.4,
  "sample_rate_hz": 48000,
  "bit_depth": 16,
  "eu_ai_act_doc": "annex/article-10.pdf",
  "license": "ypai-commercial-perpetual",
  "delivery": "2026-04-28T14:30:00Z"
}

03 Output

Dataset & docs

You receive a versioned dataset bundle plus an Article 10 compliance manifest, a QA report, and per-file metadata. Everything needed to ship to procurement, legal, and your ML team at the same time.

Versioned dataset
Article 10 manifest
QA report (Kappa, WER)
Per-file metadata
Speaker-consent ledger

PROOF

“We needed Nynorsk and Jutlandic Danish at production volume in 8 weeks. ypai delivered both with audit-ready documentation. Our WER on Western Nordic dialects dropped 12.4% after retraining.”

Head of Speech AI

Tier-1 European ASR provider

300+ million end-users · EU-headquartered
“We replaced 19% of agent-routed calls with self-service after retraining on ypai’s regional dialect data. Procurement signed off because the audit trail mapped directly to our DPIA. Six markets, six languages, one contract.”

VP, Conversational AI

European telecommunications group

12 countries · 80+ million subscribers
“Statutory minority-language access used to mean a backlog of manually-transcribed citizen submissions. ypai’s Sámi and Frisian coverage let us automate intake without dropping accuracy below our 92% acceptance threshold. Audit-ready by default mattered as much as the coverage.”

Director of Digital Services

Northern European public-sector authority

5.4 million citizens served

Read the full case studies (under NDA — available on request)

TIER-1 ASR
EU TELCO
AUTOMOTIVE OEM
PUBLIC SECTOR
EU PUBLISHER
NORDIC BANK

Logos shown by category. And 200+ other EU enterprises. Specific references available under NDA in scoping call.

COMPLIANCE

Audit-ready from day one.

Every dataset arrives with the documentation your procurement team will ask for and your auditor will verify. Built into our pipeline, not bolted on.

Download Article 10 documentation template

EU AI Act effective Aug 2026 · Our datasets are audit-ready today.

quickstart.py | python

from ypai import Client

client = Client(api_key="ypai_sk_live_...")
dataset = client.datasets.get("ypai-NO-nyn-2026Q2")
manifest = dataset.download("./local/path", with_audit=True)

print(f"Downloaded {manifest.hours:.1f}h, audit doc at {manifest.audit_path}")

quickstart.sh | bash

curl https://api.ypai.ai/v1/datasets/ypai-NO-nyn-2026Q2 \
  -H "Authorization: Bearer $YPAI_API_KEY" \
  -H "X-Include-Audit: true" \
  --output ypai-NO-nyn-2026Q2.tar.gz

REST + signed URLs

Every dataset is fetchable with a 15-min signed URL — stream straight to your S3, GCS, Azure, or HTTPS endpoint.

EU data residency

Frankfurt or Stockholm hosting, enforced per contract. SCC + DPA available. Speaker-consent ledger included.

Audit endpoint

GET /v1/datasets/{id}/audit returns the EU AI Act Article 10 documentation as JSON or PDF.

Voice data for the languages everyone else skips.

The languages and dialects we cover.

Norwegian Nynorsk

Jutlandic Danish

Swiss German

Galician

Sámi

Frisian

Welsh

Corsican

Your specification, our pipeline, your data.

Brief & spec

Pipeline, end to end

Dataset & docs

Audit-ready from day one.

EU AI Act Article 10

GDPR Articles 5, 6, 9, 22

ISO 27001 + SOC 2 Type II

EU Data Residency

Five lines to your first dataset.

REST + signed URLs

EU data residency

Audit endpoint

Scope your dataset in 30 minutes.