TTS and voice cloning data, recorded by voices you choose
We record text-to-speech and voice cloning data with native speakers reading your script as sentence lists or monologues, and you can pick the voices from short samples. Every file is checked before transcription, and every speaker’s consent is documented.
From voice sample to transcribed recordings
You pick the voices
If you want to choose the voices, each candidate speaker records a short sample. You accept the ones that fit your product, and only those speakers go on to record.
Your script
Speakers record your sentence or prompt list, or monologues. Dialogue between two speakers is possible too, for conversational voices.
Checked before transcription
Microphones and recording locations are tested before the first session. Every file then goes through automatic checks for hardware problems, background noise and static, and files that pass are reviewed by a native speaker.
What a project can include
| Voices | Native speakers to your quotas for age, gender, region and accent; you can approve each voice from a recorded sample |
|---|---|
| Speech | Sentence or prompt lists, monologues, or dialogue between two speakers |
| Recording | On our platform, by speakers whose identity, hardware and background noise are checked first |
| Transcripts | Human-verified in your convention, verbatim or clean-read |
| Consent | A briefing on how the recordings will be used, a speaker contract for the project, and explicit commercial-use consent, documented |
| Delivery | Audio, transcripts and per-speaker metadata in the layout your pipeline expects |
| Languages | Any of our 60 catalog languages and dialects, and others after a feasibility check |
Built for
- Single-speaker TTS voices
- Voice cloning and custom voices
- Conversational voices for voice agents
- TTS in languages with little public data
Catalog or custom?
Our off-the-shelf datasets are licensed for TTS training, but they are two-person conversations, not one voice reading a script. A chosen voice is recorded to your spec. For what a TTS corpus needs, read our guide to TTS datasets.
Frequently asked questions
Where can I get TTS training data recorded with a specific voice?
SpeechData.ai records TTS and voice cloning data to order with native speakers, who read your sentence lists, prompts or monologues on our platform. You can approve each voice from a short sample before recording starts. You get a feasibility assessment and a quote within one business day of sending your brief.
How is consent handled for voice cloning data?
Before recording, each speaker watches a briefing that explains what the recordings will be used for, then reads the contract for that specific project and chooses to sign or drop out. Every speaker gives explicit commercial-use consent, and the documentation is delivered with the data for your legal review.
Can your off-the-shelf datasets be used to train TTS?
Yes. The license covers training and evaluating ASR, TTS and other voice AI models. The catalog is spontaneous two-person conversation from 7,250 native speakers, with one channel per speaker, so it suits multi-speaker and conversational models rather than one chosen voice.
Which languages can you record?
Any of the 60 languages and dialects in our catalog, and other languages on request. We confirm recruitment feasibility for each language before quoting.
Describe the voices your model needs
Share your languages, how many voices, speaker profiles, script and hours per voice. We’ll come back within one business day with a feasibility assessment and a quote.