DOVEXAI/XIMA-2122

DOVEXAI Nigerian Languages Speech & Translation Dataset

Curated by DOVEXAI LTD (Nigeria) · dovexai.africa · dovexai.io

A provenance-tracked dataset of Nigerian-language voice recordings paired with human transcriptions and translations, built for supervised fine-tuning (SFT) of speech and translation models. Every recording is cryptographically receipted, fully anonymized, and quality-gated before release.

---

Dataset snapshot

Feature	Detail
Split	train (20 examples)
Audio format	M4A, mono
Recording style	Code-switched: Nigerian English → Local language in one clip
Languages	Hausa · Nigerian English · Nigerian Pidgin
Language distribution	Hausa ▉▉▉▉▉▉▉▉▉▉ 50% · Nigerian English ▉▉▉▉▉ 25% · Nigerian Pidgin ▉▉▉▉▉ 25%
Contributor gender	Female 100%
Provenance	Cryptographically receipted, verified manifest
License	CC BY 4.0

Every recording in this release is sourced from a female contributor.

How to use

from datasets import load_dataset

ds = load_dataset("DOVEXAI/XIMA-2122")

# Play/access audio
sample = ds["train"][0]
print(sample["audio"])          # {'array': [...], 'sampling_rate': 44100, 'path': ...}
print(sample["text_en_ng"])     # English prompt
print(sample["text_local"])     # Local-language transcription
print(sample["language"])       # e.g. "Hausa"

Dataset columns

Column	Type	Description
id	string	Unique sample identifier
audio	Audio	The voice recording (playable in the Dataset Viewer)
text_en_ng	string	The Nigerian-English prompt that was spoken
text_local	string	Verbatim transcription of exactly what the contributor said in the recording, including natural/colloquial spelling. This is the form anchored to the audio.
text_local_standard	string	(optional) A standardized rendering of the same line. This is a normalized/"clean" version, not necessarily what is heard in the audio.
language	string	The language spoken in this clip (e.g. Hausa, Nigerian Pidgin, Nigerian English)
local_lang_key	string	Machine key for the local language (e.g. hausa, nigerian-pidgin, nigerian-english)
speech_type	string	code-switched (English then local language) or monolingual (Nigerian English only)
receipt_id	string	Links the row to its provenance receipt
audio_sha256	string	SHA-256 of the original audio file for integrity verification

Important: verbatim vs. standardized. text_local is the ground truth for what was actually spoken and should be used as the transcription label. text_local_standard, when present, is an editorial standardization and may differ from the audio.

Recording structure

Each recording falls into one of two types, indicated by the speech_type column:

Code-switched (speech_type = "code-switched")

The contributor first reads the prompt in Nigerian English, then immediately speaks the local-language translation in the same clip:

[ Nigerian English sentence ]  →  [ Hausa / Nigerian Pidgin translation ]

This applies to:

• DVX-XIMA-2122-001 (all 10 clips): English → Hausa
• DVX-XIMA-2122-003 clips 6-10: English → Nigerian Pidgin

Each clip yields a two-for-one pair: the same audio carries both text_en_ng (the English portion) and text_local (the local-language portion).

Monolingual (speech_type = "monolingual")

The contributor speaks Nigerian English only. There is no local-language translation in the clip.

• DVX-XIMA-2122-003 clips 1-5: purely Nigerian English

For these recordings, local_lang_key is nigerian-english and text_local contains the same English text.

Additional files

This dataset also ships the following raw files for advanced use cases:

sft.jsonl                          supervised pairs: audio path → text  (raw format)
audio/<receipt_id>/audio_NNN.m4a   the voice recordings (renamed, no personal info)
receipts/<receipt_id>.json         per-contributor, hash-locked provenance receipts
receipts/index.json                master manifest (tamper-evident)

---

Provenance & integrity

• Each receipt lists every source file with its SHA-256 and carries a receipt_hash computed over the receipt's contents; index.json carries an index_hash over the manifest. Recomputing these detects any tampering.
• Audio is shipped unmodified (no denoising, trimming, or normalization), so the file hash in each receipt matches the released .m4a byte-for-byte.
• Recordings passed an automated noise-floor quality gate before inclusion.

Privacy & consent

• Fully anonymized. Contributors appear only as opaque ids (contributor_NNN); real names and original filenames are never included.
• Contributors recorded under signed consent and NDA agreements. Those documents are not shipped. Each receipt proves they exist via their SHA-256 hash alone, with no personal data exposed.

License

Released under Creative Commons Attribution 4.0 International (CC BY 4.0). You are free to use, share, and adapt this dataset, including for commercial purposes and for training models, provided you give appropriate credit to DOVEXAI LTD. Full terms: https://creativecommons.org/licenses/by/4.0/

---

Citation

DOVEXAI LTD (Nigeria). DOVEXAI Nigerian Languages Speech & Translation Dataset.
dovexai.africa / dovexai.io

For questions about this dataset, contact DOVEXAI LTD via dovexai.africa or dovexai.io.