SaarAI/asr-benchmark-outputs

SaarAI ASR Benchmark Outputs

Raw per-utterance model outputs (transcription manifests) produced by the gsma-asr-bench runners on SaarAI/asr-leaderboard-datasets.

• files: 84
• utterances: 373416
• languages: 5
• models: 22

Layout

data/<language_name>/<split>__<dataset_config>__<model_slug>.jsonl
index.jsonl   # one record per file (language, split, model, rows, sha256, ...)
index.csv

Directories categorise by language name; the file name begins with the split name, so both can be parsed without opening the file:

from pathlib import Path
split, dataset_config, model_slug = Path(path).stem.split('__', 2)
language_name = Path(path).parent.name

Loading

from datasets import load_dataset

# everything
ds = load_dataset("SaarAI/asr-benchmark-outputs", data_files="data/**/*.jsonl", split="train")

# one language / one split
ds = load_dataset("SaarAI/asr-benchmark-outputs", data_files="data/amharic/ambient_room_noise__*.jsonl", split="train")

Row schema

audio_filepath, duration, time, text (reference), pred_text (hypothesis) plus the metadata added at upload time: model_slug, model_family, language, language_name, iso_639_3, dataset_config, split, benchmark_dataset.

Contents

language	splits	files	utterances
amharic	ambient_room_noise, clean_studio, distant_low_volume_mic, harsh_environment	10	159650
kinyarwanda	test	10	100760
ndebele	test	10	19330
shona	test	13	22737
zulu	test	41	70939