datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Vaani-transcription-partThis dataset is part of the Vaani dataset and consists of only transcribed speech data. It has a total duration of 2041.54 hours, covering 59 languages.
This table represents the audio and transcription duration data for various languages.
Language
Angami
Angika
Ao
Assamese
Awadhi
Bajjika
Bearybashe
Bengali
Bhili
Bhojpuri
Bundeli
Chakhesang
Chakma
Chhattisgarhi
English
Garhwali
Garo
Gondi
Gujarati
Halbi
Haryanvi
Hindi
IduMishmi
Kannada
Kashmiri
Karbi
Khariboli
Khortha
Kokborok
Konkani… See the full description on the dataset page: https://huggingface.co/datasets/ARTPARK-IISc/Vaani-transcription-part.kumawood-speech-transcriptions
Kumawood Speech Transcriptions
Speech segments from Ghanaian films, each paired with the film's human-authored
English subtitle and a machine Twi transcript.
Total number of hours 249.8 hours
Fields
field
meaning
audio
16 kHz mono FLAC segment
text
English subtitle displayed during the segment (human-authored, recovered by OCR)
twi_text
Twi transcript from Google STT (ak) — machine output
twi_words_per_sec
transcript words per second of audio… See the full description on the dataset page: https://huggingface.co/datasets/ghananlpcommunity/kumawood-speech-transcriptions.kumawood-speech-transcriptions
Kumawood Speech Transcriptions
Speech segments from Ghanaian films, each paired with the film's human-authored
English subtitle and a machine Twi transcript.
Total number of hours 249.8 hours
Fields
field
meaning
audio
16 kHz mono FLAC segment
text
English subtitle displayed during the segment (human-authored, recovered by OCR)
twi_text
Twi transcript from Google STT (ak) — machine output
twi_words_per_sec
transcript words per second of audio… See the full description on the dataset page: https://huggingface.co/datasets/ghanaopenai/kumawood-speech-transcriptions.youtube_transcriptions
Dataset Description
A speech dataset of Uzbek language audio clips sourced from YouTube videos. Audio segments were extracted, separated by speaker using vocal isolation, and transcribed using Google's Gemini 2.0 Flash model. Speaker identities were clustered using ECAPA-TDNN embeddings.
Use Cases
Automatic Speech Recognition (ASR) for Uzbek
Text-to-Speech (TTS) synthesis for Uzbek
Fine-tuning speech models on Uzbek language data (e.g., Qwen3-TTS)
Speaker-conditioned TTS… See the full description on the dataset page: https://huggingface.co/datasets/openbank-uz/youtube_transcriptions.entity-transcription-benchmark
Entity Transcription Benchmark
Measures whether a speech recognition system transcribes named entities
correctly — as distinct from word error rate.
WER weights every token equally. The tokens that matter for redaction, lookup,
routing and search are proper nouns, and they are a small fraction of any
transcript. A system can improve WER while getting worse at exactly the words a
downstream consumer needs, and nothing in the standard evaluation will show it.
2,151 clips, 6.0… See the full description on the dataset page: https://huggingface.co/datasets/modulate/entity-transcription-benchmark.potomitan-gcf-transcription
Kreyol Guadeloupe Transcription Dataset
Ce jeu de données contient des segments audio courts (~5 secondes) en créole guadeloupéen (gcf), extraits d’émissions de radio et de télévision.
Il vise à entraîner des modèles de reconnaissance automatique de la parole (ASR) pour une langue vivante mais peu disposant de peu de ressources écrites.
Dataset Description
Le créole guadeloupéen (Karukéya) est une langue créole à base lexicale française, parlée principalement en… See the full description on the dataset page: https://huggingface.co/datasets/POTOMITAN/potomitan-gcf-transcription.transcription-corpus
UN Transcription Corpus
Two splits of UN meeting audio paired with official verbatim records.
Splits
sessions — Whole meeting sessions (SC + GA plenary)
One row per meeting. Audio from UN Web TV, verbatim records from documents.un.org.
Column
Description
symbol
UN document symbol, e.g. S/PV.9826
webtv_url
URL on UN Web TV
duration_ms
Session duration in milliseconds
num_speakers
Number of speaker turns in the verbatim record
audio_floor
Floor… See the full description on the dataset page: https://huggingface.co/datasets/united-nations/transcription-corpus.doctor-patient-convers-transcriptions-PII-redactedThis dataset contains real-world transcriptions of doctor–patient conversations in English (USA accent), focused on two medical specialties: ENT (Ear, Nose, Throat) and Dermatology and Orthopaedic. All conversations were originally recorded in clinical settings and transcribed by human experts. To comply with privacy regulations, only the transcription files are released, with all personally identifiable information (PII) fully redacted.
Due to regulations, we are only able to publish… See the full description on the dataset page: https://huggingface.co/datasets/AIxBlock/doctor-patient-convers-transcriptions-PII-redacted.Audio-Transcription-Models-Comparison-PT-BR
Audio Transcription Models Comparison
A dataset dedicated to comparing the performance of modern Speech-to-Text (STT) models, focusing exclusively on Brazilian Portuguese.
About the Dataset
This dataset was created to store and compare transcription results from different Artificial Intelligence models in challenging scenarios. Unlike generic benchmarks, this project focuses on the reality of usage in Brazil, covering:
Regionalism: Local vocabulary, accents, and… See the full description on the dataset page: https://huggingface.co/datasets/tech4humans/Audio-Transcription-Models-Comparison-PT-BR.wolof_speech_transcription
Wolof Speech Transcription
Description
Dataset de reconnaissance automatique de la parole (ASR) en wolof, une langue d'Afrique de l'Ouest parlée par plus de 10 millions de locuteurs, principalement au Sénégal.
Ce dataset est un miroir HuggingFace du corpus wolof du projet ALFFA hébergé à l'origine sur GitHub par le laboratoire GETALP (Grenoble).
Source originale
Ce dataset provient du projet ALFFA :
Repository : getalp/ALFFA_PUBLIC
Laboratoire : GETALP… See the full description on the dataset page: https://huggingface.co/datasets/serge-wilson/wolof_speech_transcription.massive-yt-edu-transcriptions
Massive YouTube Educational Transcriptions
Large-scale educational content transcribed from YouTube using distil-whisper/distil-large-v3.5.
Stats
Videos: 59,355
Characters: 1,539,022,925 (~384M tokens)
Audio hours: 35,890
Model: faster-whisper (CTranslate2) with distil-large-v3.5
Hardware: 2x RTX 5090 + 2x RTX 4090 at 165-185x realtime
Fields
Field
Description
video_id
YouTube video ID
title
Video title
text
Full transcript… See the full description on the dataset page: https://huggingface.co/datasets/thepowerfuldeez/massive-yt-edu-transcriptions.Eng-Filipino-Accented-audio-with-human-transcription-call-center-topicThis dataset contains 103+ hours of spontaneous English conversations spoken in a Filipino accent, recorded in a studio environment to ensure crystal-clear audio quality. The conversations are designed as role-play scenarios between agents and customers across a variety of call center domains.
🗣️ Speech Style: Natural, unscripted role-playing between native Filipino-accented English speakers, simulating real-world customer interactions.
🎧 Audio Format: High-quality stereo WAV files, recorded… See the full description on the dataset page: https://huggingface.co/datasets/AIxBlock/Eng-Filipino-Accented-audio-with-human-transcription-call-center-topic.transcription-results
UN Transcription Benchmark Results
Evaluation results for speech-to-text systems on UN Security Council and General Assembly meeting recordings, assessed against official UN verbatim records.
See united-nations/transcription-corpus for the underlying audio and ground truth data.
Metrics
WER: Word Error Rate (reference = verbatim record, no normalization)
normalized_wer: WER after lowercasing, punctuation removal, and filler word removal
CER: Character Error Rate (same… See the full description on the dataset page: https://huggingface.co/datasets/united-nations/transcription-results.French-Medical-Transcription-Benchmark
🩺 French Medical Transcription Evaluation Dataset
Ce dataset a été créé et ouvert à la communauté dans le cadre du développement R&D de LucioleScribe, la plateforme souveraine de transcription IA 100% locale, spécifiquement conçue pour les milieux médicaux et juridiques (compatibilité RGPD, HDS, et architectures Air-Gapped).
🔗 Découvrir LucioleScribe Édition Santé | ⚙️ Voir le Pipeline Technologique Local
📊 Présentation du Dataset
L'évaluation des modèles de… See the full description on the dataset page: https://huggingface.co/datasets/AWANNABY/French-Medical-Transcription-Benchmark.transcription-scorer
Transcription Scorer Dataset
The Transcription Scorer dataset was created to support research in reference-free evaluation of Automatic Speech Recognition (ASR) systems using human feedback. Unlike traditional evaluation metrics such as WER and its derivatives, this dataset reflects judgments of ASR outputs by human raters across multiple criteria, simulating the way a teacher grades students.
⚙️ What’s Inside
This dataset contains 1200 audio samples (from diverse sources… See the full description on the dataset page: https://huggingface.co/datasets/RobotsMali/transcription-scorer.Thai-H2H-Call-center-audio-with-human-transcriptionThis dataset contains natural Thai-language conversations between human agents and human customers, designed to reflect realistic call center interactions across multiple domains. All conversations are conducted through unscripted role-playing, allowing for spontaneous and dynamic exchanges that closely mirror real-world scenarios.
🗣️ Speech Type: Human-to-human dialogues simulating customer-agent interactions.
🎭 Style: Non-scripted, spontaneous role-playing to capture authentic speech… See the full description on the dataset page: https://huggingface.co/datasets/AIxBlock/Thai-H2H-Call-center-audio-with-human-transcription.English-USA-NY-Boston-AAVE-Audio-with-transcriptionThis dataset captures spontaneous English conversations from native U.S. speakers across distinct regional and cultural accents, including:
🗽 New York English
🎓 Boston English
🎤 African American Vernacular English (AAVE)
The recordings span three real-life scenarios:
General Conversations – informal, everyday discussions between peers.
Call Center Simulations – customer-agent style interactions mimicking real support environments.
Media Dialogue – scripted reads and semi-spontaneous… See the full description on the dataset page: https://huggingface.co/datasets/AIxBlock/English-USA-NY-Boston-AAVE-Audio-with-transcription.massive-audio-transcription-pipeline
massive-audio-transcription-pipeline outputs
Transcription outputs from the
massive-audio-transcription-pipeline,
a parallel Whisper pipeline that chunks long audio into overlapping windows,
transcribes across a worker pool, merges lightweight speaker diarization, and
checkpoints every chunk for crash resume.
Generation method
Backend: faster-whisper base model (CTranslate2), 1 worker.
Audio: real public-domain speech from the Hugging Face LibriSpeech dummy… See the full description on the dataset page: https://huggingface.co/datasets/narinzar/massive-audio-transcription-pipeline.speech-transcription-samples
Speech Transcription Samples Dataset
Synthetic speech-to-text transcription samples for ASR research.