DatoricTTS Voice Dataset
Available · 30+ languages
Build expressive multilingual TTS and voice systems from natural 48 kHz conversation with transcripts, speaker metadata, pronunciation coverage, and audio quality signals.
Dataset specification
- 48 kHz WAV audio
- Coverage: 30+ languages
- Transcripts
- Diarization metadata
- Speaker attributes
- Accent and locale labels
- SNR, LUFS, clipping, and silence metrics
- Quality scores
- TTS suitability scores
- CSV/JSON metadata
Intended uses
- Text-to-speech and speech synthesis
- Voice cloning and speaker modeling
- Multilingual conversational AI and voice agents
- Pronunciation, accent, and expressive-speech modeling
Scope notes
- Language, accent, speaker, and recording distributions are confirmed for the selected delivery.
- Transcript, diarization, normalization, and pronunciation fields vary by subset and are identified in the review package.
- Repository metadata illustrates the schema; real audio is supplied in the buyer review package.
Collection scope
Speech coverage
- 30+ global, regional, and underrepresented languages
- Natural conversational dialogue
- Accent, locale, and speaker-profile metadata where available
Language mix
- English and major accents
- Major commercial languages
- High-growth regional languages
- Low-resource and long-tail languages
Available scale
- Available program
- 48 kHz WAV or PCM where available
Buyer review package
- Annotation schema, data dictionary, and sample metadata in the repository
- Real audio, transcripts, metadata, and QA summaries for review
- Target languages, accents, speaker mix, and hours confirmed for the selected delivery
Annotation & metadata fields
- Language, locale, and accent
- Speaker count and speaker attributes where available
- Transcript and diarization availability
- Topic category and noise level
- Clean and expressive speech duration
- Text-normalization and pronunciation metadata where available
- SNR, LUFS, clipping, and silence metrics
- Quality and TTS-suitability scores
Capture methodology
- Natural conversational speech recorded for expressive pacing, tone, turn-taking, and pronunciation variation
- 48 kHz audio in WAV or PCM delivery formats where available
- 16-bit or 24-bit audio depending on the selected subset
- Utterance metadata aligned with transcripts and diarization files where available
Provenance & rights chain
- Contributor consent and chain-of-custody documentation included in licensing review
- Commercial license issued directly by Datoric
- Language, locale, and speaker-profile records travel with the selected delivery where available
- Collection and annotation activity handled under Datoric's published privacy notice
Quality, duplicates & PII
How submissions are reviewed and cleaned before they are accepted into the dataset.
- SNR, loudness, clipping, silence, and background-noise checks
- Transcript and diarization quality evaluated by subset where available
- Speaker consistency, pronunciation coverage, phoneme coverage, and expressive range review
- Duplicate, malformed, and privacy-sensitive records handled under the agreed acceptance and redaction criteria
Formats & delivery
- 48 kHz WAV audio
- CSV metadata
- JSON transcripts and diarization files where available
- Repository review materials on Hugging Face; production transfer arranged directly
Rights & license scope
Licensed directly by Datoric for commercial AI training, with final scope controlled by the signed agreement for the selected delivery.
Version & verification
- Availability
- Available
- Datasheet version
- July 21, 2026
- Release date
- July 21, 2026
- Last verified
- July 21, 2026
- Owner
- Datoric
Frequently asked
- Can we review audio before selecting a subset?
- Yes. Request a buyer review package with real 48 kHz audio in your target languages, transcripts, metadata, QA summaries, and licensing documentation.
- Which languages and accents are available?
- The program covers 30+ languages across major commercial, high-growth regional, and long-tail groups. The exact language, accent, and speaker mix is confirmed for the selected delivery.
- What quality signals are included?
- Available metadata includes SNR, LUFS, clipping, silence, noise level, transcript quality, and TTS-suitability fields, with subset-specific coverage documented during review.
- Can Datoric collect a different voice specification?
- Yes. Custom programs can target languages, accents, speaker profiles, recording conditions, scripts, and acceptance criteria.