datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Emilia-Dataset
Emilia: An Extensive, Multilingual, and Diverse Speech Dataset for Large-Scale Speech Generation
This is the official repository 👑 for the Emilia dataset and the source code for the Emilia-Pipe speech data preprocessing pipeline.
News 🔥
2025/02/26: The Emilia-Large dataset, featuring over 200,000 hours of data, is now available!!! Emilia-Large combines the original 101k-hour Emilia dataset (licensed under CC BY-NC 4.0) with the brand-new 114k-hour Emilia-YODAS… See the full description on the dataset page: https://huggingface.co/datasets/amphion/Emilia-Dataset.short_video_ocr_dataset
Short Video OCR / ASR Dataset
An actively curated research dataset for building OCR, ASR, subtitle-alignment,
and video-transcript pipelines for short social videos. It combines source
videos and extracted frames with human review artifacts and model-generated
text candidates. The primary languages are Ukrainian and Russian; English or
mixed-language content may also occur.
Status: work in progress. Model outputs and pseudo-label candidates are
not ground truth. Only… See the full description on the dataset page: https://huggingface.co/datasets/ElectronicHug/short_video_ocr_dataset.Luhya-ASR-Data-subset-642H
Luhya ASR Data Subset 642H
Luhya speech dataset for automatic speech recognition.
LEMAS-Dataset-train
Overview
This dataset is part of LEMAS-Project (lemas-project.github.io/LEMAS-Project).
It contains a large-scale training set (150k+ hours) and a curated evaluation set
(500 utterances per language) covering 10 languages, all with word-level alignment.
Fields
key: unique utterance identifier; the first two characters indicate the language ID
audio: relative path to the MP3 audio file (in the eval set, this key is renamed to "file_name" for compatibility with the viewer)… See the full description on the dataset page: https://huggingface.co/datasets/LEMAS-Project/LEMAS-Dataset-train.Bagpiper_SFT_Data
Bagpiper SFT Data
Release status: the validated Parquet release is being uploaded. The
homepage and metadata may appear before every large shard is committed.
Bagpiper SFT Data is the supervised fine-tuning corpus for
Bagpiper, an open-ended audio language model
that understands and generates speech, music, environmental sound, and their
mixtures through rich textual captions and planning.
The public release has exactly two configurations:
Configuration
Direction… See the full description on the dataset page: https://huggingface.co/datasets/espnet/Bagpiper_SFT_Data.100-hour-Egyptian-dataset-single-speaker
Masri 100h — Egyptian Arabic Single-Speaker Speech Corpus
A 100-hour Egyptian Arabic (مصري) single-narrator speech collection — 15,653 released clips at 24 kHz mono, with aligned transcripts.
Egyptian Arabic is the most widely understood Arabic dialect and one of the least served by open speech data.
Almost every open Arabic corpus is Modern Standard Arabic (MSA) — a register nobody actually speaks at home.
This dataset is built for the opposite: natural, spoken, conversational… See the full description on the dataset page: https://huggingface.co/datasets/ehabnegm/100-hour-Egyptian-dataset-single-speaker.X-Voice-Dataset-Train
X-Voice Training Dataset
Overview
The X-Voice training dataset is a large-scale multilingual speech corpus curated for high-performance speech models. It provides a robust foundation for cross-lingual phonetic and prosodic modeling.
Also the train set of X-Voice Model.
Core Statistics
Total Speech Duration: 420K hours
30 languages
European: bg (Bulgarian), cs (Czech), da (Danish), de (German), el (Greek), en (English), es (Spanish), et (Estonian), fi… See the full description on the dataset page: https://huggingface.co/datasets/XRXRX/X-Voice-Dataset-Train.Somali-ASR-Subset-68H
Somali ASR Subset 68H
Somali speech dataset for automatic speech recognition.
khmer-speech-dataset
Khmer ASR Cultural Dataset
727.94 hours of manually curated speech-text pairs by native speakers in the Khmer language about Cambodian cultural topics. On average, each recording is 8 seconds. Speaker metadata (gender, age group, and origin city) is provided.
Language: Khmer (khm).
Source(s): Native speakers from Cambodia (5 females, 7 males). The utterances were manually generated based on topics and subtopics listed in metadata.
Domain(s): Cultural domain, with a total of 61… See the full description on the dataset page: https://huggingface.co/datasets/Digital-Divide-Data/khmer-speech-dataset.asr-leaderboard-datasets
ASR Leaderboard Datasets
This repository contains test splits from multiple speech corpora, including FLEURS, Common Voice (MCV), and Multilingual LibriSpeech (MLS).
How to Load
To load a specific subset, use load_dataset with the corresponding config_name in the format <set>_<lang>.
from datasets import load_dataset
# Load the FLEURS dataset for Bulgarian
fleurs_bg = load_dataset("nithinraok/asr-leaderboard-datasets", "fleurs_bg")
print(fleurs_bg)
# Load the MCV… See the full description on the dataset page: https://huggingface.co/datasets/nithinraok/asr-leaderboard-datasets.Kamba-ASR-Data-Subset-484H
Kamba ASR Data Subset 484H
Kamba speech dataset for automatic speech recognition.
ASR-datasets-ptbr
📚 Datasets de Áudio em Português (PT-BR)
Este repositório reúne diversos corpora públicos de fala em português do Brasil, combinados em um único dataset para facilitar treinamentos e pesquisas em ASR (Automatic Speech Recognition).
O objetivo é fornecer um recurso amplo, padronizado e de fácil acesso para a comunidade.
📂 Datasets Integrados
A tabela abaixo lista todos os datasets incluídos, com suas informações:
Dataset
Config Name
TOTAL
train
test
validation… See the full description on the dataset page: https://huggingface.co/datasets/opedromartins/ASR-datasets-ptbr.Bagpiper_PreTrain_Data
Bagpiper Pretraining Data
Bagpiper Pretraining Data is the public rich-captioned audio snapshot associated
with Bagpiper, an open-ended audio language
model that learns bidirectional mappings between audio and comprehensive text
descriptions across speech, music, environmental sound, and mixtures.
The en metadata describes the primary rich-caption language. Source audio can
contain speech or singing in other languages; it is not an English-only audio
guarantee.
The repository… See the full description on the dataset page: https://huggingface.co/datasets/espnet/Bagpiper_PreTrain_Data.Gusii-ASR-Data-Subset-470H
Gusii ASR Data Subset 470H
Gusii speech dataset for automatic speech recognition.
gptsovits_dataset
bhyuan/gptsovits_dataset
GPT-SoVITS speech dataset, packed as WebDataset tar shards.
Layout
data/
train/
metadata.csv
audio/
train-000.tar
train-001.tar
...
validation/
metadata.csv
audio/
validation-000.tar
...
test/
metadata.csv
audio/
test-000.tar
...
Shard counts:
youshengshu_v5_test: 6536 tar shard(s)
Inside each tar, every sample is a pair sharing a unique key:
<key>.wav # raw… See the full description on the dataset page: https://huggingface.co/datasets/bhyuan/gptsovits_dataset.SimbaBench_dataset
SibmaBench Data Release & Benchmarking
To evaluate your model on SimbaBench across all supported tasks (ASR, TTS, and SLID), simply load the corresponding configuration for the task and language you wish to benchmark.
Each task is organized by configuration name (e.g., asr_test_afr, tts_test_wol, slid_61_test). Loading a configuration provides the standardized evaluation split for that specific benchmark.Example:
from datasets import load_dataset
data =… See the full description on the dataset page: https://huggingface.co/datasets/UBC-NLP/SimbaBench_dataset.propagator-multimodal-pretraining-data
Propagator Multimodal Pretraining Data
This public dataset contains tokenized multimodal pretraining data prepared for the Propagator model family. It combines language, image-grounded, and speech/audio-token examples into a single training format.
This is not a raw text or image browsing dataset. The examples have already been converted into compact binary token frames for model training, with a manifest that records the source groups and file layout.
Source Code… See the full description on the dataset page: https://huggingface.co/datasets/ken-sungmin/propagator-multimodal-pretraining-data.linto-dataset-audio-ar-tn
LinTO DataSet Audio for Arabic Tunisian A collection of Tunisian dialect audio and its annotations for STT task
This is the first packaged version of the datasets used to train the Linto Tunisian dialect with code-switching STT
(linagora/linto-asr-ar-tn).
Dataset Summary
Dataset composition
Sources
Data Table
Data sources
Content Types
Languages and Dialects
Example use (python)
License
Citations
Dataset Summary
The LinTO DataSet Audio for Arabic Tunisian is a diverse… See the full description on the dataset page: https://huggingface.co/datasets/linagora/linto-dataset-audio-ar-tn.khm-asr-cultural
Khmer ASR Cultural Dataset
134.6 hours manually curated speech-text pairs by native speakers in Khmer language about Cambodian cultural topics. On average, each recording is 8.54 seconds with the standard deviation of 3.37. Speaker metadata (gender, age group, and origin city) is provided.
Language: Khmer (khm).
Source(s): Native speakers from Cambodia (4 females, 4 males). The utterances were manually generated based on topics and subtopics listed in metadata.
Domain(s):… See the full description on the dataset page: https://huggingface.co/datasets/Digital-Divide-Data/khm-asr-cultural.eka-medical-asr-evaluation-dataset
Eka Medical ASR Evaluation Dataset
Dataset Overview and Sourcing
The Eka Medical ASR Evaluation Dataset enables comprehensive evaluation of automatic speech recognition systems designed to transcribe medical speech into accurate text—a fundamental component of any medical scribe system. This dataset captures the unique challenges of processing medical terminology, particularly branded drugs, which is specific to the Indian context.
The dataset comprises over 3,900+… See the full description on the dataset page: https://huggingface.co/datasets/ekacare/eka-medical-asr-evaluation-dataset.khmer-speech-dataset
Khmer ASR Cultural Dataset
727.94 hours of manually curated speech-text pairs by native speakers in the Khmer language about Cambodian cultural topics. On average, each recording is 8 seconds. Speaker metadata (gender, age group, and origin city) is provided.
Language: Khmer (khm).
Source(s): Native speakers from Cambodia (5 females, 7 males). The utterances were manually generated based on topics and subtopics listed in metadata.
Domain(s): Cultural domain, with a total of 61… See the full description on the dataset page: https://huggingface.co/datasets/phonsobon/khmer-speech-dataset.Agri_STT_Benchmarking_Dataset
Agri STT Benchmarking Dataset
10,808 farmer voice queries in Hindi, Telugu and Odia, with reference transcripts, for benchmarking automatic speech recognition in agricultural contexts. The audio is included in this repository.
Every recording is a smallholder farmer speaking a question to Farmer.Chat, an AI advisory service run by Digital Green. Reference transcripts were produced by human annotators. Nothing here is read from a script or recorded in a studio, so the audio… See the full description on the dataset page: https://huggingface.co/datasets/DigiGreen/Agri_STT_Benchmarking_Dataset.Uzbek-STT-Dataset-780h
Uzbek STT Dataset (~780 hours)
A large Uzbek speech-to-text dataset for training and fine-tuning automatic
speech recognition (ASR) models such as Whisper.
Dataset summary
Language
Uzbek (uz)
Examples
122,464
Total audio
~780 hours
Clip length
up to 30 seconds each
Columns
audio, transcription
Audio
embedded in parquet, original sample rates (e.g. 44.1 kHz / 16 kHz), mono/stereo as recorded
Split
single train split (split it yourself as… See the full description on the dataset page: https://huggingface.co/datasets/Abduqayum/Uzbek-STT-Dataset-780h.DuplexOmni-Data
DuplexOmni Data
This dataset accompanies DuplexOmni: Real-Time Listening, Seeing, Thinking, and Speaking for Full-Duplex Interaction.
GitHub: MuyeHuang/DuplexOmni
arXiv: 2606.09186
Files
The root directory includes the Writer-Director source files:
inbound.director.jsonl
outbound.director.jsonl
The synthesized parquet training shards are about 9 TB in total, so uploading them is expected to be very slow. They will be uploaded incrementally; each completed… See the full description on the dataset page: https://huggingface.co/datasets/MuyeHuang/DuplexOmni-Data.open-asr-leaderboard-multilingual-datasets
ASR Leaderboard Datasets
This repository contains test splits from multiple speech corpora, including FLEURS, Common Voice (MCV), and Multilingual LibriSpeech (MLS).
How to Load
To load a specific subset, use load_dataset with the corresponding config_name in the format <set>_<lang>.
from datasets import load_dataset
# Load the FLEURS dataset for Bulgarian
fleurs_bg = load_dataset("nithinraok/asr-leaderboard-datasets", "fleurs_bg")
print(fleurs_bg)
# Load the… See the full description on the dataset page: https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-multilingual-datasets.afri-temp-data4
AfricanVoices Hausa -- Train Split
Hausa speech dataset from AfricanVoices.io.
Usage
from datasets import load_dataset
ds = load_dataset("suleiman2003/afri-temp-data4", split="train")
print(ds[0])
# {'audio': Audio(...), 'transcript': '...', 'gender': '...', ...}
Structure
Each batch is in its own subdirectory under train/:
train/
batch_1/
*.flac + metadata.csv
batch_2/
*.flac + metadata.csv
...
Audio files are FLAC format. Metadata… See the full description on the dataset page: https://huggingface.co/datasets/suleiman2003/afri-temp-data4.kupe-asr-en-data
kupe-asr-en-mini-150m — data
Two loadable configs, packed into ~20-25 bunch_*.parquet files each (Hub-quota friendly):
raw — 24 kHz mono English audio (flac bytes) + text. The encode stage reads this.
mimi — Mimi c0..c7 codes (12.5 Hz) + text. Training reads this.
Ledgers under ledger/ (data.json, mimi.json) track collected/encoded hours and resume state.
from datasets import load_dataset
ds = load_dataset("anuj-inavlabs/kupe-asr-en-data", "mimi", split="train")
ethiopian-languages-speech-dataset
Leyu Ethiopian Languages Speech Dataset
Audio recordings paired with corresponding text transcripts, collected on the Leyu Data Collection Platform — an open-source platform for crowdsourced speech data collection — for the Leyu Platform Competition, covering 4 languages: Amharic, Afaan Oromo, Sidama, Tigrinya.
Dataset Summary
Languages: Amharic (am), Afaan Oromo (om), Sidama (sid), Tigrinya (ti)
Total examples: 2750
License: CC-BY-4.0
Task categories: Automatic… See the full description on the dataset page: https://huggingface.co/datasets/Benji-fish/ethiopian-languages-speech-dataset.hashed_data
Munch Hashed Index - Lightweight Audio Reference Dataset
📖 Overview
Munch Hashed Index is a lightweight reference dataset that provides SHA-256 hashes for all audio files in the Munch Urdu TTS Dataset. Instead of storing 1.27 TB of raw audio, this index stores only metadata and cryptographic hashes, enabling:
✅ Fast duplicate detection across 4.17 million audio samples
✅ Efficient dataset exploration without downloading terabytes
✅ Quick metadata queries (voice… See the full description on the dataset page: https://huggingface.co/datasets/humair025/hashed_data.WolneLektury-TTS-Polish
WolneLektury-TTS-Polish
A large-scale, high-quality Polish speech dataset for text-to-speech and automatic speech recognition.
Data Source
Derived from Wolne Lektury (Free Readings), a Polish digital library with public domain audiobooks featuring professional voice actors.
Dataset Statistics
Metric
Value
Total samples
383,710
Total duration
997 hours
Unique narrators
1207
Male samples
294,756 (767h)
Female samples
88,945 (230h)
Average… See the full description on the dataset page: https://huggingface.co/datasets/datadriven-company/WolneLektury-TTS-Polish.