Consented speech data for low‑resource languages.
Built to your spec, starting with native speakers across the Balkans and Eastern Europe, regional dialects included: real conversations, verbatim transcripts and speaker metadata, on an exclusive or non-exclusive licence. Text and video on request; more languages follow what teams ask for.
- Audio
- WAV or FLAC, 16-bit or 24-bit, 16–48 kHz. 8 kHz telephony on request.
- Channels
- One channel per speaker for conversations, mono for read speech.
- Speech types
- Spontaneous conversation, read prompts, short commands, expressive speech.
- Transcripts
- Verbatim or clean, speaker labels, timestamps, tags for overlaps, disfluencies, noise and code-switching.
- Metadata
- Age range, gender, region, dialect, native languages, device and recording environment, in a JSON or CSV manifest.
- Quality
- Automatic checks for clipping, noise, silence and duplicates, plus native-speaker review. Targets agreed per project.