Skip to content

Datasets

Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,788 in the catalog so far, with new ones added every hour.

Data type

For which models

Find a dataset for your goal

Answered by Gemini 4 · the chat is kept only in this tab

Datasetshb777/gemini-flash-2.0-speechTop pick

Synthetic speech from Gemini Flash 2.0 (284 hours)

An audio dataset with paired texts and voiceovers in two voices from Gemini Flash 2.0 for training TTS and ASR.

  • Text → speech
  • Automatic speech recognition
  • English
277 6110K–100K rows
Datasetechonova-ai/arabic-dialect-asr-benchmark

Arabic and 8 dialects speech recognition benchmark

20 hours of annotated Arabic speech in Modern Standard Arabic and eight regional dialects for evaluating ASR systems.

  • Automatic speech recognition
  • Arabic
277 101K–10K rows🔒 on request
DatasetHumynLabs/medical-opinion-english-audio

Noisy medical audio for robustness tests

Short English audio clips featuring medical complaints with intentionally low quality for robustness testing.

  • Audio classification
  • English
272 543up to 1K rows
DatasetAlayaLab/HiveTop pick

Dataset for text-prompted sound separation

A multimodal corpus of millions of examples for training universal query-based audio separation systems.

  • Audio → audio
  • English
270 531M–10M rows
Datasetysdede/khanacademy-turkish

Turkish speech corpus from Khan Academy videos

78 hours of short audio clips with Turkish voiceovers and text for ASR and TTS training.

  • Automatic speech recognition
  • Text → audio
  • Turkish
270 3910K–100K rows
Datasethumyn-labs/Indian-Emotional-Speech-Corpus

Indian Emotional Speech: Audio and Text

Audio recordings of Indian speakers across four emotions and paragraph text for emotion classification.

  • Audio classification
  • English
251 441up to 1K rows
Datasetupb-nlp/echo

300 Hours of Romanian Speech by 300 Speakers

Crowdsourced Romanian audio corpus for training speech recognition systems.

  • Automatic speech recognition
  • Romanian
250 5100K–1M rows🔒 on request
Datasetlitagin/moe-speech

Japanese speech from visual novels

An audio corpus of Japanese anime-style speech from about 50 visual novels, roughly 600 hours.

  • Text → speech
  • Audio → audio
  • Audio classification
  • Japanese
246 285100K–1M rows🔒 on request
DatasetHumynLabs/medical-symptoms-english-audio

Noisy audio with medical symptoms

English audio recordings with a symptom phrase, intentionally noisy to test model robustness.

  • Audio classification
  • English
242 470up to 1K rows
DatasetAbijahKaj/telephony-amd-dataset

Multilingual AMD telephony audio dataset

An audio dataset for training streaming answering machine detection models in telephony.

  • English
  • French
  • Spanish
  • +1
235 51K–10K rows
DatasetHumynLabs/banking-customersupport-hinglish-audio

Hinglish banking customer support audio

Short Hinglish bank customer audio clips for ASR, emotion classification, and voice agents.

  • Audio classification
  • English
233 661up to 1K rows
DatasetHumynLabs/e-commerce-customersupport-english-audio

English e-commerce customer support audio

English customer support audio recordings for ASR, emotion classification, and voice agents.

  • Audio classification
  • English
232 440up to 1K rows
Datasetseanghay/khmer_mpwt_speech

Khmer audio and transcripts

Nearly two hours of voiced Khmer phrases from a speech synthesis mobile app.

  • Text → speech
  • Khmer
217 111K–10K rows
DatasetMohamedRashad/multilingual-tts

Multilingual text-to-speech (TTS) dataset

Between 10k and 100k audio-text pairs in six languages for text-to-speech tasks.

  • Text → speech
  • Arabic
  • English
  • Chinese
  • +9
216 5210K–100K rows
Datasetdatapointai/text-to-speech-human-preferences-315kTop pick

315 thousand human evaluations for 15 text-to-speech models

Pairwise audio comparisons from 15 TTS models in English for RLHF and quality evaluation.

  • Text → speech
  • Reinforcement learning
  • English
212 42100K–1M rows🔒 on request
Datasetfarabi-lab/kazakh-sttTop pick

Kazakh speech dataset for ASR

High-quality open-source Kazakh speech corpus with a duration of 554 hours for ASR systems.

  • Kazakh
205 12100K–1M rows🔒 on request
Datasetuvci/koumankan4dyula

Speech and translation for the Dyula language

A corpus of 15 hours of audio in West African Dyula with transcriptions and translations in French and English.

  • Automatic speech recognition
  • Translation
  • Bambara
194 1210K–100K rows🔒 on request
DatasetHumynLabs/airline-customersupport-Hinglish-audio

Noisy airline support audio

Deliberately noisy audio recordings with a Hinglish phrase to test audio model robustness.

  • Audio classification
  • English
190 438up to 1K rows
Datasetotoearth/otoSpeech-video-full-duplex-21h

Two-way English video dialogues

A dataset of 81 full-duplex video and audio conversation sessions for audio-to-audio tasks.

  • Audio → audio
  • English
168 18up to 1K rows🔒 on request
DatasetHumynLabs/banking-customersupport-english-audio

English banking support audio

A small English audio dataset featuring a bank password reset phrase for ASR and emotion classification.

  • Audio classification
  • English
166 669up to 1K rows
DatasetHumynLabs/e-commerce-customersupport-hinglish-audio

E-commerce Hinglish support audio

Small Hinglish audio recordings for ASR, emotion, and voice agents.

  • Audio classification
  • English
162 501up to 1K rows