datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
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.UltraVoice
UltraVoice: Scaling Fine-Grained Style-Controlled Speech Conversations for Spoken Dialogue Models
📝 Abstract
Spoken dialogue models currently lack the ability for fine-grained speech style control, a critical capability for human-like interaction that is often overlooked in favor of purely functional capabilities like reasoning and question answering. To address this limitation, we introduce UltraVoice, the first large-scale speech dialogue dataset… See the full description on the dataset page: https://huggingface.co/datasets/tutu0604/UltraVoice.thai-aligner-bench
Thai Aligner Bench
🚧 Development in progress.
How accurately can a forced aligner place Thai token and word boundaries in
speech? This is a self-contained benchmark: one Python file
(aligner_bench.py) plus 1,572 clips of Thai speech with frame-exact timing
ground truth. No Thai NLP stack or other code is needed — just
numpy soundfile torch torchaudio transformers.
The ground truth is what makes the dataset useful: the audio was rendered by a
TTS model whose duration predictor… See the full description on the dataset page: https://huggingface.co/datasets/wayu-ai/thai-aligner-bench.speech-translation-and-summarization
English-Centric Multilingual Audio Dataset
This dataset contains generated article and summary audio for English-centric multilingual directions.
Each direction folder contains metadata JSONL files and corresponding audio files for few_shot and test splits.
Included directions
amharic_english / english_amharic
arabic_english / english_arabic
bengali_english / english_bengali
chinese_simplified_english / english_chinese_simplified
english_english
french_english /… See the full description on the dataset page: https://huggingface.co/datasets/McGill-NLP/speech-translation-and-summarization.quran-tajweed-phonetics
The complete phonetic layer of the Quran in the riwaya of Hafs 'an
'Asim via tariq al-Shatibiyyah: 6,236 ayat, 522,475 phones, every
phone carrying its tajweed attribution: madd class with its transmitted
length range, ghunna grade, qalqalah class, tafkheem with its rank, sakt,
the seventeen sifat, and the rule that produced it.
Built and maintained by Quran Lab, a waqf building open technology in
the service of the Quran.
How it was built and verified
Indexed from the… See the full description on the dataset page: https://huggingface.co/datasets/Quran-Lab/quran-tajweed-phonetics.librivox-tracksA dataset of all audio files & corresponding sources uploaded to LibriVox before 1st November 2025.
Donate to Internet Archive, who hosts most of the audio data
Donate to Project Gutenberg, who hosts most of the source books
Be a good netizen and use proper caching & sensible rate limiting when downloading files.
November 2025 Update
book.id, section.id, and section.num are now integers instead of strings.
reader has been replaced with a readers array. No entries currently… See the full description on the dataset page: https://huggingface.co/datasets/pykeio/librivox-tracks.bengali-talkshow-audio
Bengali Talkshow Audio Dataset
A large-scale collection of 1,180 Bengali talk show audio recordings totaling 789+ hours of multi-speaker speech, sourced from Bangladeshi television talk shows and political debate programs.
Dataset Description
This dataset contains audio from Bengali-language TV talk shows, political debates, and news discussion programs from major Bangladeshi television channels. Each recording features multiple speakers engaged in discussion, making it… See the full description on the dataset page: https://huggingface.co/datasets/nymtheescobar/bengali-talkshow-audio.multichannel-meetings-10h
GroundTruth Multi-Channel Meeting Audio Dataset (10h)
Summary
This dataset contains approximately 10 hours of co-located, multi-speaker meeting recordings, each captured simultaneously via a room (built-in) microphone and individual close-talk lapel microphones worn by each participant.
Each meeting includes:
One full meeting recording (room microphone)
Individual close-talk recordings for each participant (one file per speaker)
Structured metadata describing speakers… See the full description on the dataset page: https://huggingface.co/datasets/ground-truth/multichannel-meetings-10h.teochew_wild
Teochew-Wild：首个正字标注的野外潮州话数据集
本数据集（Teochew-Wild）是从网络上发音清晰、噪声较少的音视频内容中获取的，原始音视频的数据来源为：民生新闻、潮汕讲古、地方电视节目、故事书、抖音自媒体口播等，我借鉴了Emilla提出的数据集自动处理流水线，对原始数据进行归一化、降噪和剪切（部分自动剪切效果差的使用手工修正）；
Teochew-Wild总共包括20个发音标准、念错率低的潮汕母语说话人、共12500条音频片段，包含潮州市区、汕头市区、澄海、榕江音、潮安南部等多个区域的口音，语料内容覆盖书面用语与口头用语，并同时提供正字和拼音标注，是首个公开可用、标注准确率高的潮州话数据集，主要面向语音识别和语音合成任务。
文件说明 （File Structure Explanation）
├── label_for_qwen_asr/ # 预处理标签文件夹，完全适配Qwen-ASR模型读取格式
├── README.md # 项目说明文档（本文档）… See the full description on the dataset page: https://huggingface.co/datasets/panlr/teochew_wild.massive-yt-edu-queue
Massive YouTube Educational Video Queue
Full metadata and content classification for 4,489,228 YouTube educational videos totaling 3,975,157 hours.
Description
This dataset contains metadata, content categorization, and license risk assessment for ~4.5M YouTube videos identified as potentially educational. It serves as the discovery and processing queue for the massive-yt-edu-transcriptions project, which aims to create the world's largest open educational transcript… See the full description on the dataset page: https://huggingface.co/datasets/thepowerfuldeez/massive-yt-edu-queue.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.tr-full-dataset
TR-Full_dataset
This is a merged speech dataset containing 41427 audio segments from 88 source datasets.
Dataset Information
Total Segments: 41427
Speakers: 222
Languages: tr
Emotions: neutral, angry, sad, happy
Original Datasets: 88
Dataset Structure
Each example contains:
audio: Audio file (WAV format, original sampling rate preserved)
text: Transcription of the audio
speaker_id: Unique speaker identifier (made unique across all merged… See the full description on the dataset page: https://huggingface.co/datasets/Codyfederer/tr-full-dataset.AI_Agent_Task_Dataset
🤖 Massive AI Agent Task Dataset (10.5GB)
📌 Overview
Welcome to the AI Agent Task Dataset, a massive 10.5GB procedural dataset designed for training, fine-tuning, and evaluating autonomous AI agents and LLMs.
This dataset focuses on:
Multi-step reasoning
Tool usage (APIs, frameworks, systems)
Real-world execution workflows
Perfect for building agentic AI systems, copilots, and automation models.
📑 Table of Contents
Dataset Details
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/oncody/AI_Agent_Task_Dataset.open-vi-dialog-synthetic-100h
OpenDialog Vietnamese Synthetic Dialogue 100h
Synthetic Vietnamese two-speaker dialogue for ZipVoice-Dialog experiments.
12,000 chunks
30 seconds per chunk
100.0 hours total
Each item contains S1/S2 speaker labels, turn timings, target text,
relationship, pronouns, environment, topic, mood, and source reference IDs.
Audio renderer: vLLM-Omni VoxCPM2
Audio format: mono WAV, 48 kHz, 30 seconds per chunk
This is a research dataset. Review the source/reference licensing and the… See the full description on the dataset page: https://huggingface.co/datasets/tsdocode/open-vi-dialog-synthetic-100h.Aren
ARen — Arabic/English ASR Robustness Set
Curated and published by TNSA AI.
A small, deliberately hard evaluation set for Arabic and English speech
recognition. Every clip exists in three acoustic conditions so you can measure
not just how a model scores, but how fast it falls apart as the channel
degrades.
Built because clean read-speech benchmarks stop discriminating between modern
ASR systems long before real deployments stop breaking.
Why it exists
On clean… See the full description on the dataset page: https://huggingface.co/datasets/TNSA/Aren.pseudolabel-malaya-speech-stt-train-whisper-large-v3uzbek-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.uk-pods
uk-pods - speech datasets of Ukrainian podcasts.
Preparation
Clone the dataset repository and extract the content of clips.tar.gz archive.
git clone https://huggingface.co/datasets/taras-sereda/uk-pods
cd uk-pods && tar -zxvf clips.tar.gz
To use these manifests for training/inference with NeMo [1] modify audio_filepath to absolute locations of audio files extracted in previous step.
# data_root=<clonned_repo_dir> # /home/ubuntu/uk-pods
data_root=$(realpath .)
sed -i… See the full description on the dataset page: https://huggingface.co/datasets/taras-sereda/uk-pods.err-video-news-transcribed
Transcribed ERR Video News Dataset
This dataset contains transcriptions of video news stories from Estonian National Brroacasting (https://www.err.ee/). There are around 40K stories with a total duration of around 4000 hours.
Transcriptions are generated automatically using speech recognition (gemini-3-flash-preview). Contextual biasing was used to improve ASR quality, using the textual news story about the same topic.
The WER of the transcriptions is around 5% on the average.
The… See the full description on the dataset page: https://huggingface.co/datasets/TalTechNLP/err-video-news-transcribed.bibletts-asante-twi-repaired
BibleTTS Asante Twi — Repaired Transcripts
The Asante Twi transcripts released with BibleTTS have had the
characters ɛ (U+025B) and ɔ (U+0254) stripped out. This dataset restores them.
Audio is not included. This is a drop-in replacement for the .txt files that ship with the
BibleTTS Asante Twi package, matched by clip ID.
The problem
Both are Twi vowels, and both are required by the orthography. Measured across the released
Asante Twi transcripts:
Character… See the full description on the dataset page: https://huggingface.co/datasets/danieldzikunuofmarvel/bibletts-asante-twi-repaired.test321
test321
This is a merged speech dataset containing 118 audio segments from 2 source datasets.
Dataset Information
Total Segments: 118
Speakers: 4
Languages: tr
Emotions: happy, angry, sad, neutral
Original Datasets: 2
Dataset Structure
Each example contains:
audio: Audio file (WAV format, 16kHz sampling rate)
text: Transcription of the audio
speaker_id: Unique speaker identifier (made unique across all merged datasets)
emotion: Detected emotion… See the full description on the dataset page: https://huggingface.co/datasets/Codyfederer/test321.ted-talks-in-chinese-zhongwen
TED中文 Podcast
聚焦华语地区的创意，本节目从上万个TED和TEDx演讲中，为您精选中文演讲，以及少量中文配音的经典英语演讲。演讲人包括科技和人文专家、关心当下与未来的思考者、关注挑战与探索的实践者。英雄不论出处，谁有创意谁讲。让这些演讲成为一把把钥匙，开启你的好奇心，升级你的行动力。
Focusing on creativity within the Chinese-speaking world, this program curates Chinese-language talks from tens of thousands of TED and TEDx presentations, along with a select few English classics dubbed into Chinese. Our speakers span technology and humanities experts, thinkers engaged with the present and future, and practitioners… See the full description on the dataset page: https://huggingface.co/datasets/bdx33/ted-talks-in-chinese-zhongwen.quantized-common-voice-encommon-voice-en-revoicecleaned-asr-transcriptstestpic01
EuroSpeech Dataset
Dataset Description
EuroSpeech is a large-scale multilingual speech corpus containing high-quality aligned parliamentary speech across 22 European languages. The dataset was constructed by processing parliamentary proceedings using a robust alignment pipeline that handles diverse audio formats and non-verbatim transcripts.
Dataset Summary
Languages: 22 European languages (see detailed breakdown below)
Total aligned hours: ~78,100 hours of… See the full description on the dataset page: https://huggingface.co/datasets/zyzsasasas/testpic01.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.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.freddy-testDette er et datasett som skal slettes.
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.