License summary

Plain-language summary of our standard terms. The full Data License Agreement and Evaluation Terms come with every quote; this page does not itself grant a license.

Datasets: standard license

GrantNon-exclusive, worldwide, perpetual license to use the licensed dataset version to train, fine-tune, evaluate and improve AI models, and to use and sell those models and their outputs commercially.
Not allowedReselling or redistributing the audio or transcripts; trying to identify any speaker; cloning or imitating an individual speaker's voice; using the data for surveillance or biometric identification.
ExclusivityAvailable for custom collections at a separate price and term.
ConsentEvery speaker signed consent covering AI training and evaluation, commercial use and transfer outside Taiwan. A consent and origin record ships with each file.
WithdrawalA speaker who withdraws is removed from all future deliveries. Licensees are notified and receive a replacement version; files already used in completed training runs need not be untrained.
Personal dataNames, phone numbers, addresses and ID numbers spoken aloud are removed from audio and tagged in transcripts. No contact details of speakers are delivered.
PaymentInvoice in USD or TWD, bank transfer, 30 days.
LiabilityEach side's liability is capped at the fees paid under the order, except for breach of the restrictions above.
LawLaws of Taiwan (R.O.C.); Taiwan Taipei District Court.

Private evaluations

How it worksYou give us API or model access. We run it on held-back test audio you never receive, and send a report of error rates by variety, age band, setting and code-switching.
Test-set privacyTest audio and transcripts stay with us so scores stay honest. We delete any outputs from your model after the report unless you ask us to keep them.
ConfidentialityYour results are confidential. Public leaderboard listing only with your written opt-in.

Ask for a quote