Purpose:

To collect resources of freely available datasets for machine learning (with emphasis on speech). Links are working as of January 2019.

Healthy Speech

• Read English: LibriSpeech

• Various spoken American English sentences (with various dialects) with good annotations: TIMIT corpus (partial)

• Collection of variety of languages, voluntarily collected world-wide: Voxforge

• YouTube videos (can use audio and/or visuals) of celebrities (English): VoxCeleb.

• Variety of languages: Arctic2 Database

• Datasets/ databases for a variety of research purposes/ algorithms: The Centre for Speech Technology Research

• Corpus of Spoken Dutch

• Annotated and with natural sounds (laughs, coughs, etc): Meeting Recorder Dialog Act Database

• LibriVox: free public domain audiobooks

• Free Spoken Digit Dataset

• Two types of north American dialects speaking Harvard sentences (I believe): PN/NC Corpus

Clinical Speech

• Various disorders, as well as healthy (German): Saarbrücker Voice Database

• Children with Specific Language Impairment (Czech): LANNA database

• Interviews of people with clinical disorders (anxiety, depression, post-traumatic stress disorder): DAIC-WOZ Database

Emotion in Speech

• German: EmoDB

• English: The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS)

Speech Commands

• Google, TensorFlow, and AIY collaborated to create the Speech Commands Dataset

General Sounds / Noise / Music

• Urban Sound Dataset

• Freesound.org

• Music

• Freesound Audio Tagging dataset for Kaggle competition

• For acoustic scene analysis: LITIS Rouen Audio scene dataset

General Datasets

• Github repo for machine learning datasets: Observations

• Github repo comparing ConvNet and RNNs on time sequence data: locuslab

• An amazing list of opensource datasets

Text

• Sentiment Analysis Datasets.

• Text Comprehension: SQuAD (The Stanford Question Answer Dataset)

Images

• Object class recogntion: PASCAL VOC. One needs to register but should be able to download for free

• Object class recogntion: COCO (common objects in context)

• IMAGENET: “an image database organized according to the WordNet hierarchy”.

Pre-Trained Models

• Speech related: Kaggle Discussion.

• WaveNet can be included in ur projects I believe.