datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
AudioMarathon
🎵 AudioMarathon: A Comprehensive Benchmark for Long-Context Audio Understanding and Efficient Inference in Multimodal LLMs
Abstract
AudioMarathon is a large-scale, multi-task audio understanding benchmark designed to systematically evaluate audio language models' capabilities in processing and comprehending long-form audio content. It provides a diverse set of 10 tasks built upon three pillars:
long-context audio inputs with durations ranging from 90.0 to 300.0… See the full description on the dataset page: https://huggingface.co/datasets/Hezep/AudioMarathon.Real-TurnTurk
Real-TurnTurk
English: Real-TurnTurk is a multimodal, two-channel Turkish dyadic conversation dataset built to improve turn-taking prediction in voice-based dialogue systems. Unlike Syn-TurnTurk, the other dataset we built, every conversation here is a real, unscripted exchange between two people, recorded over video calls. Each participant was captured on a separate audio channel, so speaker attribution is exact and requires no diarization model. Alongside the audio, the… See the full description on the dataset page: https://huggingface.co/datasets/tugrulbayrak/Real-TurnTurk.Neapolitan-Spoken-Corpus
Neapolitan Spoken Corpus (NSC)
A corpus of read Neapolitan speech for ASR evaluation, with a validated
Neapolitan–Italian lexicon, LOSO fine-tuning splits, trained LoRA adapters,
metric implementations, per-clip results, and error annotations.
This release supersedes the earlier 141-clip single-speaker version of this
repository. The earlier release corresponds to Speaker S1 of the present
corpus; the old audioData/ and transcripts.csv are replaced by
data/audio/ and… See the full description on the dataset page: https://huggingface.co/datasets/michaelcacioli/Neapolitan-Spoken-Corpus.knesset-plenums
About
This dataset is derived from raw a/v recordings and human-generated protocols of the Knesset (the Israeli house of representatives) plenums as part of the ivrit.ai project.
Consider visiting the preview space for this dataset here
Method
Data dumps from the Knesset contain A/V recordings, alongside proprietary protocols with timestamps.
We extract the audio stream, and clean up timestamp mistakes (such as backward jumps, or out-of-order timestamp artifacts).
The… See the full description on the dataset page: https://huggingface.co/datasets/ivrit-ai/knesset-plenums.fama-data
Dataset Description, Collection, and Source
The FAMA training data is the collection of English and Italian datasets for automatic speech recognition (ASR) and speech translation (ST)
used to train the FAMA models family.
The ASR section of FAMA is derived from the MOSEL data collection, including the automatic
transcripts obtained with Whisper and available in the HuggingFace MOSEL Dataset.
The ASR is further augmented with automatically transcribed speech from the… See the full description on the dataset page: https://huggingface.co/datasets/FBK-MT/fama-data.maleo-short-1.5H
Dataset Card for Maleo Short 1.5H
Dataset Description
Dataset Summary
Maleo Short 1.5H is a manually curated, rigorously annotated speaker diarization dataset designed to benchmark State-of-the-Art (SOTA) models against complex, "in-the-wild" media domains. While modern diarization pipelines excel in controlled acoustic environments (like telephony or reading corpora), they heavily struggle with the overlapping speech, sound effects, and rapid speaker shifts… See the full description on the dataset page: https://huggingface.co/datasets/maleo-ai/maleo-short-1.5H.VoiceCommandAudioThis is mainly used for fine tune "VoiceCommand" a speech congnition MOD dedicated for SilentHunter game series
modicol
MoDiCoL - A Modular Diagnostic Continual Learning Dataset for ASR
MoDiCoL is a speech dataset designed to study the robustness of ASR models to different drift factors in a controlled, continual setting. We construct MoDiCoL using a systematic factorial design that enables a rigorous evaluation of linguistic, speaker, and acoustic variation with clearly defined experimental runs. By combining real-world and synthetic speech with a configuration-dependent augmentation pipeline… See the full description on the dataset page: https://huggingface.co/datasets/TPekarekRosin/modicol.pao-audio-dataset
🎙️ Pa'O Audio Dataset
ပအိုဝ်ႏ အငေါဝ်း အဆင်ႏဗာႏ ရွမ်ခြွဉ်းဗူႏ
📌 Project Summary
The Pa'O Audio Dataset is an open-source initiative created to facilitate the development of speech technologies and Artificial Intelligence tools for the Pa'O language (ပအိုဝ်ႏဘာႏသာႏငေါဝ်းငွါ).
Pa'O is primarily spoken in Shan State and other regions of Myanmar. As a low-resource language in the AI landscape, this dataset provides audio recordings and corresponding… See the full description on the dataset page: https://huggingface.co/datasets/paodigitalhub/pao-audio-dataset.clipquill-asr-benchmark
Measuring whisper-tiny vs whisper-base in a browser tab
Word error rate, wall-clock timing, transfer size and peak memory for two
quantised Whisper tiers running entirely client-side in a real Chrome window,
with the scripts that produced every number.
If you are building an in-browser transcription page, the two results worth
knowing before you pick a model tier:
On clean synthetic audio the two tiers tie. If that is all you test, you
will conclude the tier does not matter… See the full description on the dataset page: https://huggingface.co/datasets/sophia8888/clipquill-asr-benchmark.open-ko-s2s-eval-artifacts
Open Ko-S2S 평가 산출물 (감사용)
⚠️ KsponSpeech 참조 전사는 해시로 대체돼 있습니다
KsponSpeech 는 AI Hub 배포 데이터로 재배포 제한이 있을 수 있어, kspon 런의
ref 컬럼을 ref_sha256 으로 대체했습니다(전사 원문 미포함). 모델 출력(hyp)과
채점 결과(cer_err/cer_len/cer)는 우리 산출물이라 그대로 공개합니다.
Zeroth 런은 원본이 CC BY 4.0(OpenSLR #40)이라
ref 원문을 그대로 담고 있습니다.
라이선스 보유자의 검증 절차
AI Hub 에서 KsponSpeech 를 정당하게 받은 분은 다음으로 우리 수치를 검증할 수 있습니다.
리더보드 저장소의 eval/datasets_ko.py 에서 clean_kspon() 을 가져옵니다.
자기 사본의 원 전사에 clean_kspon() 을 적용합니다. 결과가 목록이면… See the full description on the dataset page: https://huggingface.co/datasets/baryonlabs/open-ko-s2s-eval-artifacts.parler-tts_mls_eng_10k_snac_token_old
Dataset Card for Dataset Name
This dataset card aims to be a base template for new datasets. It has been generated using this raw template.
Dataset Details
Dataset Description
Curated by: [More Information Needed]
Funded by [optional]: [More Information Needed]
Shared by [optional]: [More Information Needed]
Language(s) (NLP): [More Information Needed]
License: [More Information Needed]
Dataset Sources [optional]
Repository: [More… See the full description on the dataset page: https://huggingface.co/datasets/blanchon/parler-tts_mls_eng_10k_snac_token_old.africanvoices-naija-batch1-summary
African Voices Naija Train Metadata Summary
This dataset contains a compact summary of metadata for the Naija training split, provided as CSV tables for inspection and analysis.
Files included:
batch_summary.csv
domain_distribution.csv
The repository contains metadata summaries only and does not include raw audio.
Moroccan-Arabic-Multimodal-Emotion-Recognition
MDER-MA — Moroccan Arabic Multimodal Emotion Recognition (TTS-aligned repackaging)
A repackaging of the MDER-MA dataset that pairs every audio clip with its Arabic (Moroccan dialect / Darija) transcript and ships speaker-disjoint train/validation/test splits.
Original dataset: Ouali, S. & El Garouani, S. (2025). MDER-MA: A multimodal dataset for emotion recognition in low-resource Moroccan Arabic language. Data in Brief. DOI: 10.1016/j.dib.2025.112005. Mendeley:… See the full description on the dataset page: https://huggingface.co/datasets/FatimahEmadEldin/Moroccan-Arabic-Multimodal-Emotion-Recognition.polish-tedx-asr-eval
Polish-TEDx-ASR-Eval
A dataset for evaluating automatic speech recognition (ASR) systems for Polish in the domain of TEDx public talks.
Contains audio segments from Polish TEDx talks available on YouTube (CC BY-NC-ND 4.0) and synthetic speech generated with KugelAudio (MIT), with manually created and cross-verified transcriptions. Created as part of the course "Workshops on Evaluation of Speech Recognition Systems" (ZWESUI, AMU 2026) by Group 1.
Statistics… See the full description on the dataset page: https://huggingface.co/datasets/s512757/polish-tedx-asr-eval.script-fidelity-benchmark
Script fidelity benchmark
Anonymous supplement for the paper "Script collapse in multilingual ASR:
A reference-free metric and 100-pair benchmark."
Script Fidelity Rate (SFR) measures the fraction of ASR hypothesis characters
that belong to the expected target script. WER measures word edits, while SFR
checks whether the output is written in the target orthography.
Related resources:
PyPI package: https://pypi.org/project/script-fidelity/
Hugging Face Evaluate metric:… See the full description on the dataset page: https://huggingface.co/datasets/themechanism/script-fidelity-benchmark.BLURB-synth
BLURB-synth: Synthetic audio data based on BLURB corpora
Dataset Summary
Synthetic audio data based on BLURB corpora. More details coming soon...
Supported Tasks and Leaderboards
Biomedical Language Understanding and Reasoning Benchmark (BLURB)
Text-to-Speech
Automatic-Speech-Recognition
Languages
English
Data Structure
Data Instances
Coming soon...
Data Fields
Coming soon...… See the full description on the dataset page: https://huggingface.co/datasets/uy-rrodriguez/BLURB-synth.Saudilang-Code-Switch-Corpus
SCC - Saudilang Code-Switch Corpus
The National Center for Artificial Intelligence at the Saudi Data and Artificial Intelligence Authority (SDAIA), published the "SCC" dataset, which stands for "Saudilang Code-Switch Corpus”.
This dataset contains a transcription of general conversations taken from a YouTube podcast "Thmanyah" that has been transcribed by the National Center for Artificial Intelligence in SDAIA. The data features three episodes covering different domains: investment… See the full description on the dataset page: https://huggingface.co/datasets/SDAIANCAI/Saudilang-Code-Switch-Corpus.dwesui-grupa-1-neurologia
NeuroSpeechPL
Publiczny eksport HuggingFace zawiera wyłącznie redystrybuowalne audio source=natural. Wiersze TTS są celowo wyłączone z publicznego zbioru danych, ponieważ ich source_license zabrania redystrybucji audio. Pełna lokalna ewaluacja opisana w raporcie korzystała zarówno z nagrań naturalnych, jak i TTS.
Repozytorium zbioru danych HF: https://huggingface.co/datasets/JankesTNJ/dwesui-grupa-1-neurologia
Repozytorium kodu:… See the full description on the dataset page: https://huggingface.co/datasets/JankesTNJ/dwesui-grupa-1-neurologia.zamai-pashto-voice2voice
ZamAI Pashto Voice2Voice
This dataset contains Pashto voice-to-voice preparation metadata for speech and translation experiments. It focuses on Pashto speech records, dialect information, transcript text, and a small viewer-ready sample manifest.
Configs
from datasets import load_dataset
metadata = load_dataset("ZamAI-Pashto/zamai-pashto-voice2voice", "metadata")
sample = load_dataset("ZamAI-Pashto/zamai-pashto-voice2voice", "viewer_sample")
Files… See the full description on the dataset page: https://huggingface.co/datasets/ZamAI-Pashto/zamai-pashto-voice2voice.ascend
Dataset Card for ASCEND
Dataset Summary
ASCEND (A Spontaneous Chinese-English Dataset) introduces a high-quality resource of spontaneous multi-turn conversational dialogue Chinese-English code-switching corpus collected in Hong Kong. ASCEND consists of 10.62 hours of spontaneous speech with a total of ~12.3K utterances. The corpus is split into 3 sets: training, validation, and test with a ratio of 8:1:1 while maintaining a balanced gender proportion on each set.… See the full description on the dataset page: https://huggingface.co/datasets/filwsyl/ascend.Shinekhen-BuryatAudio collected by Yamakoshi (Tokyo University of Foreign Studies), originally uploaded here (CC BY-SA 4.0).
start_time and end_time are from the original audio clips; the audio uploaded here are already converted into per-sentence audio clips.
Used in [paper] [GitHub]
lost-in-speech
Lost in Speech
A trilingual benchmark for reference-free classification of synthetically introduced factual and contextual alterations in English, Russian, and Kazakh. It contains 12,013 samples derived from news articles, with text, synthesized speech, and ASR transcript representations used in the study.
Altered samples are LLM-generated rewrites with a controlled alteration type—contradiction, fabrication, or context inconsistency—and severity level—mild, moderate, or severe.… See the full description on the dataset page: https://huggingface.co/datasets/maristombayeva/lost-in-speech.ben10-asr-results
Ben-10 Regional ASR — public results
Score rows for the maintainer-run Ben-10 regional dialect ASR leaderboard.
Field
Meaning
model_id
Hub id or slug
model_url
Link to weights / paper
wer
Corpus Word Error Rate on private ben-10-test (lower better)
wer_by_region
JSON map region → WER
backend
Decode stack used by maintainers
scorer_commit / decode_commit
Git SHAs in BengaliAI/reg-speech-aacl
evaluated_at
ISO date
requested_by
Who asked, or maintainer if… See the full description on the dataset page: https://huggingface.co/datasets/bengaliAI/ben10-asr-results.Arabic-Emotional-Audio-Dataset-Baved
BAVED — Basic Arabic Vocal Emotions Dataset (TTS-ready repackaging)
A re-packaged, transcript-aligned version of the Basic Arabic Vocal Emotions Dataset (BAVED) with explicit Arabic transcripts, English glosses, speaker metadata, and speaker-disjoint train/validation/test splits.
Original dataset: Aouf Yacine, Basic Arabic Vocal Emotions Dataset (BAVED), GitHub: https://github.com/40uf411/Basic-Arabic-Vocal-Emotions-Dataset. This repackaging adds metadata; all audio is unchanged.… See the full description on the dataset page: https://huggingface.co/datasets/FatimahEmadEldin/Arabic-Emotional-Audio-Dataset-Baved.zamai-pashto-voice2voice
ZamAI Pashto Voice2Voice
Languages: psLicense: cc-by-4.0Task categories: automatic-speech-recognition, audio-to-audioSize categories: n<1K
Summary
This dataset is part of the ZamAI Pashto data collection. It is intended for automatic-speech-recognition, audio-to-audio tasks in Pashto.
How to use
from datasets import load_dataset
dataset = load_dataset("tasal9/zamai-pashto-voice2voice")
print(dataset)
Configs
default: load with… See the full description on the dataset page: https://huggingface.co/datasets/tasal9/zamai-pashto-voice2voice.Swedia-ASR-Dataset
Swedia ASR Dataset
This repository contains a small Swedish ASR evaluation dataset based on speech
transcriptions from Swedia 2000. It was assembled to compare automatic
speech-recognition output against manually corrected reference transcriptions
for Swedish dialectal speech.
The dataset is useful for quick experiments with Swedish ASR systems, especially
when you want to inspect recognition quality on spontaneous speech from
different regions, speakers, ages, and genders.… See the full description on the dataset page: https://huggingface.co/datasets/kvest/Swedia-ASR-Dataset.luel-multilingual-tts-samples
Multilingual TTS Samples (Luel)
License: All Rights Reserved. Proprietary. Access only for authorized parties; no redistribution or use without permission. See LICENSE.
A multilingual text-to-speech / read-speech dataset of short scripted utterances across 7 languages. Each sample is a single-speaker recording of a written prompt, paired with rich speaker and recording metadata. Useful for TTS training and evaluation, ASR adaptation, dialect/accent studies, and read-speech… See the full description on the dataset page: https://huggingface.co/datasets/Luel-ai/luel-multilingual-tts-samples.crowd-whatsapp-yi
About
This dataset was created by crowd-sourced Whatsapp voice recordings in Yiddish as part of the ivrit.ai project.
Volunteers read a message sent to them from a predefined set of messages, recording themselves using Whasapp voice message sent to the collecting bot.
Later this data is normalized by aligning the captions with the audio using Stable Whisper (See Below).
The recording project is an ongoing effort and new data will be appended to this dataset periodically as it is… See the full description on the dataset page: https://huggingface.co/datasets/ivrit-ai/crowd-whatsapp-yi.crowd-recital-yi
About
This dataset was created by crowd-sourced recording sessions in Yiddish as part of the ivrit.ai Crowd Recital project.
Volunteers read on normal desktop or mobile setting Wikipedia articles while time-stamping every sentence read.
Later this data is normalized by aligning the gathered captions with the audio using Stable Whisper (See Below).
The recording project is an ongoing effort and new data will be appended to this dataset periodically as it is being generated.… See the full description on the dataset page: https://huggingface.co/datasets/ivrit-ai/crowd-recital-yi.