CoolFace
Datasetpublic

Arozo/aiza-malagasy-administrative-asr

AIZA — Malagasy/French Code-Switched Administrative Speech (Benchmark Set) 10 short, self-recorded, consented audio clips of Malagasy/French code-switched questions about Malagasy administrative procedures (lost ID card, birth certificate, passport, land title, etc.) — created as the benchmark audio for AIZA, a submission to the Sahara CodeSwitch Africa Main Challenge. Why this dataset exists Neither Sahara's published language list nor the official… See the full description on the dataset page: https://huggingface.co/datasets/Arozo/aiza-malagasy-administrative-asr.

sourceHugging Facecc-by-nc-4.0updated 6d agoView on Hugging Face
0likes84downloads
Dataset Card

AIZA — Malagasy/French Code-Switched Administrative Speech (Benchmark Set)

10 short, self-recorded, consented audio clips of Malagasy/French code-switched questions about Malagasy administrative procedures (lost ID card, birth certificate, passport, land title, etc.) — created as the benchmark audio for AIZA, a submission to the Sahara CodeSwitch Africa Main Challenge.

Why this dataset exists

Neither Sahara's published language list nor the official intronhealth/AfriSwitch code-switching benchmark covers Malagasy. To honestly benchmark speech recognition on AIZA's actual target domain, we recorded our own small, consented test set rather than claim results on a language pair no existing public benchmark covers.

Consent and privacy

  • Every clip is self-recorded by the dataset creator, specifically for this benchmark, with full awareness of its public use here.
  • Content is limited to generic administrative example sentences (no real names, ID numbers, or personal data of any kind).

Files

  • audio/<id>.webm — 10 short recordings (see metadata.csv for per-file duration)
  • metadata.csv — id, topic, reference transcript, duration (seconds), languages
  • ground_truth.json — same reference transcripts, JSON form
  • results.json — full benchmark results (Sahara + 2 Malagasy-specific ASR models: WER/CER per utterance) — see the submission's Benchmark Report for the human-readable writeup

Stats

  • 10 utterances, ~44 seconds total
  • Languages: Malagasy (mg) code-switched with French (fr)
  • Domain: civic/administrative queries

License

CC BY-NC 4.0 — free to use for non-commercial research and benchmarking, with attribution.