datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
esb-datasets-earnings22-validation-tiny-filteredA filtered (<=30s duration) slice (512 samples) of the Earnings22 dataset.
def add_duration(sample):
y, sr = sample['audio']["array"], sample['audio']["sampling_rate"]
sample['duration_ms']=librosa.get_duration(y=y, sr=sr) * 1000
return sample
tedlium = load_dataset("esb/datasets", "earnings22", split='validation', trust_remote_code=True)
# compute duration to filter
tedlium = tedlium.map(add_duration)
tedlium = tedlium.select(range(512))
# Whisper max supported duration
tedlium… See the full description on the dataset page: https://huggingface.co/datasets/D4nt3/esb-datasets-earnings22-validation-tiny-filtered.tr-subset-v0.1
TR Subset v0.1 — Turkish speech
251,118 Turkish audio/text rows (~62 GB, 128 parquet shards). Schema is just
text + audio; see the YAML header above.
An early-phase Turkish speech collection from the TinyAya data pipeline. It is
not part of the v0.3 Stage-2 training corpus — that is
tr-hi-mimi-encoded.
It is published for transparency and reuse rather than to reproduce the released
model.
from datasets import load_dataset
ds = load_dataset("tiny-aya-translate/tr-subset-v0.1"… See the full description on the dataset page: https://huggingface.co/datasets/tiny-aya-translate/tr-subset-v0.1.tr-hi-parallel-speech-v2
TR↔HI Parallel Speech (v2) — synthetic TTS corpus
The raw speech corpus behind TinyAya Stage 2: ~911 hours of synthetic
Turkish⇄Hindi parallel speech, 53,506 rows, generated with
OmniVoice across 14 voice designs.
This is the pre-encoding source. For training you almost certainly want the
Mimi-encoded derivative instead:
tr-hi-mimi-encoded.
Layout
path
contents
data/train-*.parquet
the loadable table (schema in the YAML header above)
audio/*.wav
~9… See the full description on the dataset page: https://huggingface.co/datasets/tiny-aya-translate/tr-hi-parallel-speech-v2.librispeech15k_augm_train-tiny
Dataset Card for "librispeech15k_augm_train-tiny"
More Information needed
TinyStress-15K
📚 TinyStress-15K Dataset
TinyStress-15K is a synthetic dataset developed as part of our paper: "WhiStress: Enriching Transcriptions with Sentence Stress Detection". It is designed to support research of models that understand sentence stress i.e., emphasis on specific words that affect sentence meaning.
Check out our project page to access more resources.
📦 Dataset Summary
Name: TinyStress-15K
Type: Synthetic speech dataset with stress annotations
Samples: 15… See the full description on the dataset page: https://huggingface.co/datasets/slprl/TinyStress-15K.everyayah_curated_1s_20s_balanced_tinyhinglish-casual
Hinglish Casual Speech
33,275 casual Hindi-English code-switched utterances (~31 GB) with audio,
transcripts in both Devanagari and Latin script (utterance /
utterance_latin), speaker ids, style metadata and durations. Full schema is in
the YAML header above.
Collected during the TinyAya programme to probe code-switched speech, which
neither the FLORES-derived text nor the TTS corpora cover. It is not part of
the v0.3 Stage-2 training set — that is
tr-hi-mimi-encoded.
from… See the full description on the dataset page: https://huggingface.co/datasets/tiny-aya-translate/hinglish-casual.freesound-laion-640k-commercial-16khz-tiny
About this Repository
This repository is the training split of the complete FreeSound LAION 640k dataset, limited only to licenses that permit commercial works, resampled to 16khz using torchaudio.transforms.Resample.
This is ideal for use cases where a variety of audio is desired but fidelity and labels are unnecessary, such as background audio for augmenting other datasets.
Dataset Versions
The full dataset contains 403,146 unique sounds totaling 37.5 GB.
The large… See the full description on the dataset page: https://huggingface.co/datasets/benjamin-paine/freesound-laion-640k-commercial-16khz-tiny.whisp_tiny_spot_data_alltiny-humminglibrispeech_augm_validation-tiny
Dataset Card for "librispeech_augm_validation-tiny"
More Information needed
librispeech_tiny
Dataset Card for "librispeech_tiny"
More Information needed
Tiny_Processedwhisper_transcriptions.reazonspeech.tinycv-tr-eval
Common Voice Turkish Eval
4,825 Turkish test clips (~45 MB, 16 kHz) in the Mozilla Common Voice schema:
transcription, duration, up_votes / down_votes, and the age / gender
/ accent speaker attributes. Schema in the YAML header above.
An evaluation-only Turkish counterpart to
lahaja-eval;
never trained on. Used to sanity-check Turkish ASR quality on real human
speech, which matters here because the v0.3 training corpus is entirely
synthetic TTS and the released model is… See the full description on the dataset page: https://huggingface.co/datasets/tiny-aya-translate/cv-tr-eval.persian-traditional-instruments-tinyasr-common-tinycodec-superb-tinylahaja-eval
LAHAJA Hindi ASR Eval
3,076 Hindi test utterances (~712 MB) carrying rich speaker metadata —
native_language, native_state, gender, age_group, scenario — plus both
verbatim and normalized transcripts. Schema in the YAML header above.
Held as an evaluation set only: never trained on. Its dialect and
native-state labels make it useful for checking whether Hindi ASR quality holds
across accents rather than only on the average.
This is the benchmark behind hindi-tts-probe, which… See the full description on the dataset page: https://huggingface.co/datasets/tiny-aya-translate/lahaja-eval.tiny-hummingtiny_common_voice_16.1Tiny_Train_Processedkotoba-whisper-tiny-dataset
Dataset Card for "kotoba-whisper-small-dataset"
More Information needed
CAESAR-TINY
Dataset Card for CAESAR-TINY
Dataset Summary
CAESAR-TINY is a synthetic code-switched dataset generated by combining monolingual samples in Catalan and Spanish.
The process includes trimming silences, normalizing audio volume, and introducing random pauses. It contains 2 hours of speech data, created by concatenating audio from the Common voice 17 Benchmark split and VoxForge Spanish datasets.
Example Usage
To load CAESAR-TINY:
from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/CAESAR-TINY.vslp2020_vinai_100h_vi_pseudo_labelled_tinyeval-whisper-tiny-eka-hard-20260408-1921
Evaluation Results: whisper-tiny
Evaluation results from Whisper model evaluation.
Summary
Model
WER
CER
openai/whisper-tiny
139.85%
78.04%
Source Data
Evaluation Dataset: Trelis/eka-hard
Model Evaluated: openai/whisper-tiny
Columns
Column
Description
audio
Audio sample (if available from source dataset)
reference
Ground truth transcription
prediction
Model prediction
wer
Word Error Rate for this sample
cer… See the full description on the dataset page: https://huggingface.co/datasets/Trelis/eval-whisper-tiny-eka-hard-20260408-1921.common-voice-scripted-speech-kab-26-tiny
Common Voice Scripted Speech 26.0 - Kabyle (Cleaned)
This is a cleaned, speaker-disjoint subset of Mozilla Common Voice 26.0 for Kabyle (Taqbaylit) ASR.
Source
Original: Mozilla Common Voice 26.0 (cv-corpus-26.0-2026-06-12)
Dataset ID: cmqim4fux00tynq07ljtyhzfh (Mozilla Data Collective)
License: CC0-1.0
Generated: 2026-07-12
Cleaning Pipeline
Step
Input
Output
Filter
Quality filter
609,940
573,073
≥2 upvotes, 0 downvotes
Character… See the full description on the dataset page: https://huggingface.co/datasets/boffire/common-voice-scripted-speech-kab-26-tiny.dataset_for_review_tiny
UNO-Bench Tiny (Sample Dataset)
This is a representative sample of the full UNO-Bench dataset, created for reviewer inspection per NIPS submission guidelines (for datasets larger than 4GB).
How the Sample Was Created
The sample was generated by random sampling from the full dataset validation split:
Source: validation.parquet from the full UNO-Bench dataset (3,730 rows total).
Sampling method: Random sampling without replacement, using random_state=42 for… See the full description on the dataset page: https://huggingface.co/datasets/blue-tundra-42/dataset_for_review_tiny.gigaspeech-tiny-stage1arc_whisper_transcriptions.reazonspeech.tiny
Dataset Card for "arc_whisper_transcriptions.reazonspeech.tiny"
More Information needed