The parent site's footer names Spire Light ApS, CVR 44262509, in Valby, Denmark, while no consent text, license agreement or terms of service is published.
SpeechData.ai sells data rather than buying it, and DataPacts records no program through which a company or speaker can sell recordings to it. Every figure here is SpeechData.ai's own statement with its check date, and the profile follows the same method as every vendor profile in an independent reference on AI training data licensing.
What SpeechData.ai Sells: Off-the-Shelf and Custom Speech Datasets
SpeechData.ai sells two things to AI developers: finished conversational speech datasets from its own catalog and speech collected to a customer's specification. The offerings it names include these (checked 2026-11-09):
- Off-the-shelf conversational speech datasets.
- Custom speech collection.
- Call center data, with no price stated.
- Text-to-speech and voice data.
- Wake word data.
The buyers are AI developers training speech recognition, speech synthesis and conversational models, and free samples are available on request.
How organizations license recorded speech in general, and the specifications buyers check, are set out on the page about selling speech and audio data for AI training.
SpeechData.ai Company Identity and Operator
SpeechData.ai is operated within Spirelight, and the parent site's footer names the company behind it as Spire Light ApS, CVR 44262509, Ingrid Marievej 49, 2500 Valby, Denmark (checked 2026-11-09). The table shows what each site states.
| Attribute | Value | Source |
|---|---|---|
| Brand | SpeechData.ai, "a Spirelight company" | speechdata.ai |
| Operator named on the parent site | Spire Light ApS | spirelight.ai footer |
| Register number | CVR 44262509 (Denmark), from the parent footer; the CVR register itself was not opened | spirelight.ai footer |
| Address | Ingrid Marievej 49, 2500 Valby, Denmark | spirelight.ai footer |
| Website | speechdata.ai | speechdata.ai |
| Type | data vendor (catalog seller) | speechdata.ai |
| Founder | Andreas Kromann Bjærg, as named on SpeechData.ai's own page | speechdata.ai |
| Founded | not recorded | n/a |
| Contracting entity in published terms | none published | speechdata.ai |
The evidence chain has two links: speechdata.ai's footer names only "SpeechData.ai" and calls the site "a Spirelight company", and the parent site's footer names Spire Light ApS. No published terms name the party that signs a dataset license, so a buyer confirms the contracting entity in the license itself.
SpeechData.ai's home page names Andreas Kromann Bjærg as founder, a self-description, and no founding date is recorded. Danish rules for a data holder sit under the rules for licensing data for AI training in Denmark.
SpeechData.ai Terms at a Glance, Dated
SpeechData.ai publishes its prices, catalog size and consent statement on its home page but no license agreement or terms of service; the table lists what it states, as of 2026-11-09. Each value is SpeechData.ai's statement.
| Term | What SpeechData.ai states | Checked | Source |
|---|---|---|---|
| What is sold | off-the-shelf conversational speech datasets and custom collection | 2026-11-09 | speechdata.ai |
| Price | USD 60 to 95 per hour of audio, depending on language | 2026-11-09 | speechdata.ai |
| Samples and discounts | free samples; volume and multi-language discounts on request | 2026-11-09 | speechdata.ai |
| License wording | "cleared for commercial use" | 2026-11-09 | speechdata.ai |
| Speaker consent | "every speaker signed an explicit commercial-use consent" | 2026-11-09 | speechdata.ai |
| Compliance documentation | offered for GDPR and EU AI Act reviews; not published | 2026-11-09 | speechdata.ai |
| License agreement or terms of service | none published | 2026-11-09 | speechdata.ai |
| Seller program, payout or referral fee | none on record | 2026-11-09 | speechdata.ai |
| Speaker pay | not disclosed | 2026-11-09 | speechdata.ai |
The table is the buyer-side view, because SpeechData.ai sells datasets and publishes no seller program, payout or referral fee.
Terms that are not published, including the license agreement, permitted uses beyond "commercial use", redistribution, warranties and governing law, have to be read in the license a buyer is offered.
Programs that pay data holders publish terms of a different kind, and those sit side by side in the table of payouts and program terms compared.
SpeechData.ai Per-Hour Prices as Stated
SpeechData.ai states that its speech datasets cost USD 60 to 95 per hour of audio, depending on language, with its English (US) set of 1,500 hours and 150 speakers at USD 90 per hour (checked 2026-11-09). The table gives the figures it names.
| Dataset or scope | Price per hour of audio | Size stated | Checked | Source |
|---|---|---|---|---|
| Catalog range | USD 60 to 95, "depending on language" | 73k hours, 7,250 speakers, 60 languages and dialects | 2026-11-09 (wording 2026-11-16) | speechdata.ai |
| English (US) | USD 90 | 1,500 hours, 150 speakers | 2026-11-09 | speechdata.ai |
| A Spanish dataset | USD 65 | not recorded | 2026-11-09 | speechdata.ai |
| Call center data (service) | no price stated | not recorded | 2026-11-09 | speechdata.ai |
| Other catalog datasets | no per-dataset price recorded | not recorded | 2026-11-09 | speechdata.ai |
List prices before discounts; not what speakers or holders are paid.
The range is one catalog range, per hour of audio and depending on language, and it is not a call center price. Volume and multi-language discounts are on request, so a quoted price can differ from the list.
A list price is what a buyer pays, not what a speaker or a holder of recordings receives, and speaker pay is not disclosed, so no margin is computed here. Other per-hour figures are collected under AI training data prices per hour, where units differ and are not averaged.
What SpeechData.ai's Speaker Consent Statement Covers and Leaves Open
SpeechData.ai states that "every speaker signed an explicit commercial-use consent" and calls its datasets "cleared for commercial use", which is a claim about its own collection that the site does not document publicly (checked 2026-11-09). These items are not published:
- The consent form text.
- Whether the consent names AI training or only commercial use.
- Withdrawal and deletion after delivery.
- Speaker pay.
- The license that passes rights to the buyer.
Voice recordings of identifiable speakers are personal data in the EU, so a buyer checks the legal basis and notices described under GDPR and AI training data. Whether a voice is also biometric data is a separate question.
Before signing, a buyer can ask for the consent template, the speaker count for each dataset and the documentation SpeechData.ai says it offers.
Compliance Documentation SpeechData.ai Offers Its Customers
SpeechData.ai states that it hands its customers documentation so that GDPR and EU AI Act reviews do not stall procurement, but the documentation itself is not public (checked 2026-11-09).
The obligation sits with the model provider, not the vendor. Regulation (EU) 2024/1689 requires a copyright policy under Art. 53(1)(c) and a public training content summary under Art. 53(1)(d), applying from 2025-08-02, enforced by the Commission from 2026-08-02, and for models placed before 2025-08-02 by 2027-08-02. What the Article 53 training data summary and copyright policy require is covered on its own page.
Under the Commission template C(2025) 8311 of 2025-12-05, a licensed dataset is disclosed only as Yes or No plus modality (Section 2.2.1), so the vendor's name need not appear in a summary.
For a buyer, the practical point is to keep the vendor's documentation and the license on file.
Can a Data Holder or Speaker Sell to SpeechData.ai?
No program for companies or speakers to sell recordings appears on SpeechData.ai's site as of 2026-11-09: it sells datasets and states no price it pays for audio. SpeechData.ai is one of several vendors that sell their own datasets and state no seller program, so "no seller program on record" is the accurate wording.
No contributor program is named on SpeechData.ai's page. Pay for recording and transcription work is compared under AI training jobs and platforms compared.
A company holding recordings finds the routes that license from holders on the speech page linked above, and the call recordings page in the vendor comparison below.
SpeechData.ai Reputation: The Record as of 2026-11-09
No complaint, incident, court case or regulator action involving SpeechData.ai appears in a primary or named-person source checked as of 2026-11-09. Directory rankings that appear in search results are not evidence.
The only documented gap is disclosure: no terms and no consent text are published. Documented buyer incidents are logged in the dated list of data buyer incidents, which holds no SpeechData.ai entry.
Benefits, Risks, Downsides and Alternatives When Licensing From SpeechData.ai
For an AI developer, SpeechData.ai offers published per-hour prices and a stated consent basis but no public license or consent text, so the practical alternatives are other speech catalogs, open corpora and not licensing at all.
Benefits (as SpeechData.ai states them, checked 2026-11-09)
- Per-hour prices are published, with an English (US) example at USD 90.
- A consent basis is stated for every speaker.
- The catalog spans 60 languages and dialects, and free samples are offered.
- An operator is named with a Danish CVR number on the parent site.
Risks
- The consent text, withdrawal terms and license agreement are not public.
- The contracting entity is not named in published terms.
- EU compliance documentation is offered but not shown.
Downsides
- Per-dataset prices beyond the two examples are not recorded here and discounts are on request, so the list is a starting point.
- No route exists for holders or speakers to sell to SpeechData.ai.
Alternatives (alphabetical, each a party's own claim)
- David AI: a data license agreement with terms not published (checked 2026-11-09).
- Gengo Shigen Kyokai: JPY 33,000 per resource for non-members (checked 2026-11-09).
- Linguistic Data Consortium: for-profit membership at USD 34,000 Standard or USD 40,000 Subscription (checked 2026-11-09).
- Mozilla Common Voice: release 27.0 with 295 locales under CC0 (Mozilla, as of 2026-09-11).
- Nexdata: states 1000+ companies served (as of 2026-11-09).
- Not licensing speech data at all and collecting it in house.
None of these is a recommendation. A holder weighing a sale reads the alternatives to selling data, which include keeping the recordings.
SpeechData.ai Among Speech Dataset Vendors
SpeechData.ai is one of six catalog sellers among the dataset vendors profiled here, next to distributors of licensed corpora and open speech releases. The parent page compares dataset vendors and data labs, and the hub answers who buys AI training data. A holder deciding on a route reads about choosing a buyer for your data.
- Linguistic Data Consortium, a distributor of speech corpora.
- Gengo Shigen Kyokai (GSK), the Japanese distributor.
- Nexdata, another catalog seller with speech sets.
- DataOcean AI, a speech data vendor.
- Defined.ai, a marketplace that also lists speech datasets.
- David AI, an audio dataset vendor.
- Mozilla Common Voice, the open speech corpus.
- Holders of archives read about call recordings for AI training, and individuals about selling your voice to AI.
- Language counts are put in context under multilingual and low-resource language data.
- The legal side is covered under biometric data in AI training, voiceprints included, and customer consent before licensing data for recordings a company made itself.
More About SpeechData.ai
Frequently Asked Questions About SpeechData.ai
How much does SpeechData.ai charge for speech data?
SpeechData.ai states USD 60 to 95 per hour of audio, depending on language. English (US) is USD 90 per hour (checked 2026-11-09), and discounts are on request.
Who operates SpeechData.ai?
SpeechData.ai calls itself a Spirelight company. The parent site's footer names Spire Light ApS, CVR 44262509, in Valby, Denmark (checked 2026-11-09).
Did SpeechData.ai's speakers consent to AI training?
SpeechData.ai states that every speaker signed an explicit commercial-use consent. The consent text is not published, so whether it names AI training is not shown.
Does SpeechData.ai buy recordings from companies?
No seller program, application page or price paid for recordings appears on SpeechData.ai's site as of 2026-11-09. It sells datasets to AI developers.
How large is the SpeechData.ai catalog?
SpeechData.ai states 73k hours of audio from 7,250 native speakers in 60 languages and dialects (checked 2026-11-09).