Rare-Language Conversational Speech Corpus
Spontaneous, real-world conversational speech in four rare Indian languages — transcribed, segmented, and ready for ASR and speech-LLM training.
- Rights-cleared
- Consent documented
- 🇮🇳 India data residency
- Languages
- Santali, Gondi, Tulu, Bhojpuri
- Scale
- 260 hrs · 1,850 speakers · 1850 records
- Inter-annotator agreement
- α = 0.85
- QA pass rate
- 97.90%
Measured, audited, reproducible
- IAA / Krippendorff α
- 0.85
- QA pass rate
- 97.90%
- Word error rate
- 9.10%
Transcripts are produced via a two-pass native-speaker workflow with a held-out WER of 9.1% on spontaneous speech. Segment-boundary and speaker-label agreement is α = 0.85, with a 97.9% QA pass rate. Balanced for gender and age band.
Audio is collected in the field from consenting native speakers under fair-pay contributor agreements granting commercial reuse. Each recording carries speaker demographics (consented), region, and a consent reference; all data is stored and processed within India.
Field teams capture natural dialogue across diverse acoustic settings, then segment, diarise, and transcribe with native speakers. A second reviewer corrects transcripts and a sampling audit verifies WER and label quality before each versioned release.
What's inside each record
A representative schema for Rare-Language Conversational Speech Corpus. The full data card ships the complete field dictionary, value ranges and annotation rubric.
| Field | Type | Example |
|---|---|---|
| clip_id | string (uuid) | "spk_4421…" |
| audio_path | string (wav, 16kHz) | "clips/00421.wav" |
| transcript | string (verbatim) | "ᱟᱞᱮ ᱥᱮᱛᱟᱜ ᱨᱮ ᱪᱟᱞᱟᱜ…" |
| duration_sec | float | 7.42 |
| speaker_meta | object | {age_band, gender, region} |
| language | string (ISO 639) | "san" |
| annotator_id | string (hashed) | "anr_7f3…" |
| qa_status | enum | "passed" |
| consent_ref | string | "cns_2024_…" |
Representative schema — exact fields and value ranges are documented in the dataset card shipped with every licence.
License tiers
Choose the tier that matches your use case. Every tier ships with the full data card, provenance log, and quality report.
Commercial Non-Exclusive
Contact us
Production speech licence with versioned drops and balanced demographic splits.
Request accessEnterprise / Sovereign Exclusive
Contact us
Exclusive licensing plus additional rare languages and bespoke acoustic conditions to spec.
Talk to salesRelated datasets
-
GeneralGeneral Audio / Speech 🇮🇳 Sovereign
Audiobook & Long-Form Narration — Sustained Expressive
Sustained, expressive long-form narration for audiobook-grade TTS across P0 and P1 languages.
- Languages:
- English, Hindi (P0) + 9 P1 Indic languages
- Size:
- ≥10 hrs/speaker (up to 100)
Rights-clearedLicence from
Contact us
-
GeneralGeneral Audio / Speech 🇮🇳 Sovereign
Child-Voice Expressive Speech Set (Major Indian Languages)
Consented child-voice speech in major Indian languages — expressive, read and spontaneous, with code-switching — for age-appropriate ASR, TTS and edtech voice models.
- Languages:
- Hindi, Tamil, Telugu, Bengali, Marathi, Kannada
- Size:
- 180 hrs · 2,400 child speakers
- IAA:
- 0.86
Rights-clearedLicence from
Contact us
-
GeneralGeneral Audio / Speech 🇮🇳 Sovereign
Frontend & Conversational Voice — English + Hindi (Studio)
Natural, warm, neutral studio speech for assistant-style TTS in English and Hindi, including code-mixing — the P0 priority profile.
- Languages:
- English, Hindi (P0)
- Size:
- ≥10 hrs/speaker (up to 100)
Rights-clearedLicence from
Contact us
Explore more General data
Ready to license Rare-Language Conversational Speech Corpus?
A senior data PM will scope access, residency, and licensing terms and respond within one business day.