datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
darija-asr-corpus
Darija ASR Corpus (dataset-core)
Arabizi (Latin-script) transcriptions of Moroccan Darija speech, produced for a
Whisper fine-tuning pipeline (paper not yet published -- citation forthcoming).
This repo contains four source subsets: DODa, DVoice, Wiki, and
YouTube. Each subset carries its own upstream license/terms -- see below --
because they are drawn from four different original projects.
Subsets
Config
Rows
Audio bundled?
Upstream license
Upstream source… See the full description on the dataset page: https://huggingface.co/datasets/abnajlae/darija-asr-corpus.ark-asr-3b-open-asr-leaderboard-results
ARK-ASR-3B Open ASR Leaderboard Results
Raw JSONL manifests for AutoArk-AI/ARK-ASR-3B on the public English
short-form hf-audio/open-asr-leaderboard splits.
These manifests were generated on a local 8x RTX 4090 machine and scored with
the shared Open ASR Leaderboard scorer:
PYTHONPATH=. python - <<'PY'
from normalizer.eval_utils import score_results
score_results(
'ark_asr/results.AutoArk-AI-ARK-ASR-3B_20260622_official',
'AutoArk-AI/ARK-ASR-3B',
)
PY
Important:… See the full description on the dataset page: https://huggingface.co/datasets/Edge0/ark-asr-3b-open-asr-leaderboard-results.quran-asr-husary
Quran ASR — Husary Muallim Dataset
Description
This dataset contains Quran recitation audio files by Sheikh Mahmoud Khalil Al-Husary at 16 kHz sampling rate, with Arabic transcriptions including diacritics.
Dataset Structure
Audio files: Stored in audio/ folder (e.g., audio/001_001.wav)
Data file: manifest.json (NeMo format)
Columns:
audio_filepath: Path to audio file
text: Arabic transcription with diacritics
duration: Audio duration in seconds
speaker:… See the full description on the dataset page: https://huggingface.co/datasets/NightPrince/quran-asr-husary.Tech-Sentences-For-ASR-Training
TechVoice Dataset
Work in Progress – This dataset is actively being expanded with new recordings.
Dataset Statistics
Metric
Current
Target
Progress
Duration
38m 43s
5h 0m 0s
██░░░░░░░░░░░░░░░░░░ 12.9%
Words
10,412
50,000
████░░░░░░░░░░░░░░░░ 20.8%
Total Recordings: 205 samples
Total Characters: 74,312
A specialized speech dataset for fine-tuning Automatic Speech Recognition (ASR) models on technical and developer vocabulary. Contains human-recorded… See the full description on the dataset page: https://huggingface.co/datasets/danielrosehill/Tech-Sentences-For-ASR-Training.red_ace_asr_error_detection_and_correction
RED-ACE
Dataset Summary
This dataset can be used to train and evaluate ASR Error Detection or Correction models. It was introduced in the RED-ACE paper (Gekhman et al, 2022).
The dataset contains ASR outputs on the LibriSpeech corpus (Panayotov et al., 2015) with annotated transcription errors.
Dataset Details
The LibriSpeech corpus was decoded using Google Cloud Speech-to-Text API, with the default and video models.
The word-level confidence was enabled… See the full description on the dataset page: https://huggingface.co/datasets/google/red_ace_asr_error_detection_and_correction.uzbek-asr-curated-701h
Uzbek ASR Curated Dataset (701 hours)
A curated multi-source Uzbek speech dataset for automatic speech recognition (ASR) training and evaluation.
Dataset Description
Language
Uzbek (Latin script with okina ʻ)
Total utterances
337,920
Total duration
~701 hours
Audio format
16 kHz mono WAV (PCM_16)
Manifest format
NeMo JSONL
Splits
train (94%) / val (3%) / test (3%)
Splits
Split
Utterances
Hours
Train
317,655… See the full description on the dataset page: https://huggingface.co/datasets/uzinfocom-edu-ai/uzbek-asr-curated-701h.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: 508
utterances: 4390208
languages: 7
models: 47
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… See the full description on the dataset page: https://huggingface.co/datasets/SaarAI/asr-benchmark-outputs.hebrew-asr-vn
Hebrew ASR three-source training dataset
Derived from ivrit.ai crowd-transcribe-v5, crowd-recital and knesset-committees.
Original data and transcripts are credited to ivrit.ai and its contributors.
Pinned revisions and preparation rules are in metadata/sources.json and
metadata/preparation-config.json. VoxKnesset is excluded by user decision.
Source/split
Clips
Hours
crowd-recital/test
1,557
1.071
crowd-recital/train
45,372
33.258
crowd-recital/validation
1,033… See the full description on the dataset page: https://huggingface.co/datasets/benderrodriguez/hebrew-asr-vn.darija-asr-benchmark-6speaker
Darija ASR 6-Speaker Benchmark
A fixed, paired 20-utterance benchmark read identically by 6 held-out speakers (3
female: F1, F2, F3; 3 male: M1, M2, M3 -- none present in any training corpus),
used to evaluate cross-speaker generalization for a Moroccan Darija (Arabizi)
Whisper fine-tuning pipeline (paper not yet published -- citation forthcoming).
Consent and anonymization
Written informed consent was obtained from all six speakers for the recording and… See the full description on the dataset page: https://huggingface.co/datasets/abnajlae/darija-asr-benchmark-6speaker.uzbek-asr-train-manifests
Uzbek ASR Training Manifests
The exact training, validation and test splits behind
rustam1221/uzbek-asr-gigaam:
974 hours of Uzbek speech drawn from seven public corpora, filtered, text-normalized,
and split by speaker.
No audio is copied. Each row is a pointer — a parquet file plus a row
index in the upstream dataset — and the training dataloader decodes the audio
when the batch is built. That keeps the whole corpus definition at 200 MB
instead of roughly a terabyte of… See the full description on the dataset page: https://huggingface.co/datasets/rustam1221/uzbek-asr-train-manifests.parakeet-hindi-asr
Parakeet Hindi-English Bilingual ASR
Fine-tuning NVIDIA Parakeet TDT 0.6B for bilingual Hindi-English automatic speech recognition.
Quick Start
# Download
pip install huggingface_hub
huggingface-cli download ketav/parakeet-hindi-asr --repo-type dataset --local-dir ./parakeet-hindi-asr
# Install dependencies
pip install nemo_toolkit[asr] bitsandbytes sentencepiece
# Train (after updating paths in config)
cd parakeet-hindi-asr/scripts
python ft_0.6B_hi_v3.py… See the full description on the dataset page: https://huggingface.co/datasets/ketav/parakeet-hindi-asr.asr-evaluationsKambaBench-ASR
KambaBench-ASR
Status: v0.0 — scaffold. No evaluation audio or gold transcriptions have been finalized yet.
An open, leakage-controlled, reproducible evaluation benchmark for Kamba (Kikamba, kam) automatic speech recognition (ASR).
KambaBench-ASR is designed to provide a common evaluation standard for Kamba speech-recognition systems. The benchmark is intended to be model-agnostic: any Kamba ASR system, whether based on Whisper, MMS, Omnilingual ASR, Parakeet, or another… See the full description on the dataset page: https://huggingface.co/datasets/lawmaluki/KambaBench-ASR.quranic-asr-cloud-rawdata
Quranic ASR Provider Benchmark Results
Professional benchmark artifacts for comparing commercial and official ASR providers on the Quranic ASR benchmark hosted at Quran-Lab/quranic-asr-benchmark.
This repository contains metadata, normalized result tables, raw provider responses, unchanged run scripts, scoring outputs, Tarteel streaming probes, and reports. It does not duplicate the source audio.
What Is Included
Area
Path
Purpose
Benchmark split… See the full description on the dataset page: https://huggingface.co/datasets/Quran-Lab/quranic-asr-cloud-rawdata.cleaned-asr-transcriptssynthetic-asr-vi
Synthetic ASR data — vi
Generated by Valsea-ASR/synthetic-data-pipeline.
Audio is synthetic (TTS), targeted as training data for downstream ASR finetuning.
Total audio: 50.7 hr across short (5s) and long (30s) length buckets,
each in clean and augmented variants.
Loading
from datasets import load_dataset
ds = load_dataset("<org>/synthetic-asr-vi", "short_clean")
print(ds["train"][0]["audio"]) # {"array": np.ndarray, "sampling_rate": 16000, "path": "..."}… See the full description on the dataset page: https://huggingface.co/datasets/silvermango9927/synthetic-asr-vi.cleaned-asr-transcripts-hinglish
cleaned-asr-transcripts-hinglish
bingbangboom/cleaned-asr-transcripts-hinglish is a parallel corpus containing 14k+ pairs of raw-synthetic Hindi ASR (Automatic Speech Recognition) transcripts mapped to their clean, properly punctuated, and transliterated "Hinglish" (Romanized Hindi) counterparts.
This dataset is specifically designed for ASR post-processing, transliteration models, and fine-tuning Large Language Models (LLMs) to understand and generate high-quality, conversational… See the full description on the dataset page: https://huggingface.co/datasets/SaiyanSai/cleaned-asr-transcripts-hinglish.cleaned-asr-transcripts-hinglish
cleaned-asr-transcripts-hinglish
bingbangboom/cleaned-asr-transcripts-hinglish is a parallel corpus containing 14k+ pairs of raw-synthetic Hindi ASR (Automatic Speech Recognition) transcripts mapped to their clean, properly punctuated, and transliterated "Hinglish" (Romanized Hindi) counterparts.
This dataset is specifically designed for ASR post-processing, transliteration models, and fine-tuning Large Language Models (LLMs) to understand and generate high-quality, conversational… See the full description on the dataset page: https://huggingface.co/datasets/bingbangboom/cleaned-asr-transcripts-hinglish.ts_asr_test
ts_asr_test：目标说话人 ASR 测试集（manifest-only）
ts_asr_test is a 3,928-clip (~8.7 h) Chinese/English target-speaker ASR test set, released
manifest-only: the repo ships no audio, only an audio-free recipe and a self-contained,
deterministic rebuild script. Bring your own copies of the public source corpora and run
rebuild_ts_asr_test.py to regenerate every clip bit-for-bit.
数据集简介
每条样本由一段目标说话人语音、一段同说话人的注册音频（enrollment），以及若干干扰说话人语音与背景噪声按固定配方混合而成。任务：在给定 enrollment… See the full description on the dataset page: https://huggingface.co/datasets/Boxp/ts_asr_test.nb-asr-qwen3whisperxagreement-v1
nb-asr-qwen3whisperxagreement-v1
Word-level forced alignment training data for Norwegian speech, produced by keeping only examples where two independent aligners — WhisperX and Qwen3 (Lunde forced aligner) — agree within a tight tolerance.
Dataset Description
This dataset contains 702,067 speech segments drawn from the NB-ASR Norwegian audio corpus. Each record pairs an audio file with a word-level forced alignment in a format suitable for training a… See the full description on the dataset page: https://huggingface.co/datasets/NbAiLab/nb-asr-qwen3whisperxagreement-v1.synthetic-asr-zh
Synthetic ASR data — zh
Generated by Valsea-ASR/synthetic-data-pipeline.
Audio is synthetic (TTS), targeted as training data for downstream ASR finetuning.
Total audio: 61.2 hr across short (5s) and long (30s) length buckets,
each in clean and augmented variants.
Loading
from datasets import load_dataset
ds = load_dataset("<org>/synthetic-asr-zh", "short_clean")
print(ds["train"][0]["audio"]) # {"array": np.ndarray, "sampling_rate": 16000, "path": "..."}… See the full description on the dataset page: https://huggingface.co/datasets/silvermango9927/synthetic-asr-zh.synthetic-asr-hi
Synthetic ASR data — hi
Generated by Valsea-ASR/synthetic-data-pipeline.
Audio is synthetic (TTS), targeted as training data for downstream ASR finetuning.
Total audio: 63.5 hr across short (5s) and long (30s) length buckets,
each in clean and augmented variants.
Loading
from datasets import load_dataset
ds = load_dataset("<org>/synthetic-asr-hi", "short_clean")
print(ds["train"][0]["audio"]) # {"array": np.ndarray, "sampling_rate": 16000, "path": "..."}… See the full description on the dataset page: https://huggingface.co/datasets/silvermango9927/synthetic-asr-hi.