Buy speech and audio data for AI training

What is speech and audio training data?

Speech and audio training data is recorded sound with the metadata a model needs to learn from it: transcripts aligned to time, speaker labels, language and recording conditions. Sources include broadcast and podcast archives, commissioned recordings, meeting recordings and call recordings.

What do AI teams use speech data for?

AI teams use speech data to train models that recognize speech, generate voices and understand audio. Each use needs different material.

Language coverage is a real gap. Meta's No Language Left Behind project, which covers 200 languages including 150 low-resource ones, states that reliable translation data is not available for many languages. See multilingual text data for the text side.

How is speech and audio data priced?

Speech and audio data is priced by the hour for commissioned or catalog recordings, and by the archive for company recordings. Mercor's published listing puts call centers at $20,000 to $250,000 for recorded customer conversations from operations with 50 to 5,000 agents, as compiled by ddiy.co in October 2026.

Ask whether a quote covers audio only or audio with verified transcripts, and whether consent records come with it. Our guide to AI training data pricing models compares per-hour and flat-fee structures.

Speech data raises consent and biometric questions that text does not. Illinois' Biometric Information Privacy Act lists a voiceprint as a biometric identifier. Section 15(b) requires a private entity to inform the person in writing and receive a written release before collecting one, and section 15(c) bars an entity holding biometric identifiers from selling, leasing, trading or otherwise profiting from them.

California's CCPA defines biometric information, when used or intended to be used to establish identity, to include voice recordings from which a voiceprint can be extracted, and treats biometric information processed to uniquely identify a consumer as sensitive personal information. If you plan to train voice cloning or speaker identification models, agree the license terms for those uses separately.

Recording law matters too. Federal law at 18 USC 2511(2)(d) permits a recording where one party to the communication consents, but California Penal Code section 632 makes it an offense to record a confidential communication without the consent of all parties. Recordings must have been lawfully made before they can be lawfully licensed.

If you hold UK or EU recordings

UK GDPR Article 4 defines biometric data as personal data from specific technical processing that allows or confirms the unique identification of a person. Under Article 9, biometric data processed to uniquely identify someone is special category data, which needs an additional condition for processing.

Where does licensed speech data come from?

Licensed speech data comes from broadcasters and podcast publishers, from companies that record meetings or calls, and from commissioned collections where speakers are recruited and paid to record. Commissioned data gives the cleanest consent position, because the recording was made for training from the start.

How should speech data be delivered?

Speech data should be delivered as lossless WAV or FLAC files, each with a time-aligned transcript, a speaker ID and metadata on language, accent, recording device and environment. Consent records should link to speaker IDs rather than names, so the dataset itself carries no direct identifiers.

Ask the seller to state the sample rate and channel layout in the data manifest, along with total hours by language and the number of speakers. Long stretches of silence or music inflate the hours without adding training value.

What should you check before buying speech data?

Before buying speech data, ask for consent records linked to speaker IDs and listen to a random sample against its transcripts. Check for music, broadcasts or other third-party audio in the background, which needs separate clearance.

Our due diligence checklist covers the contract, and our guide to licensing video and audio archives explains what sellers prepare. To source vetted speech and audio, book a free consultation.