A phonebook for speech, music, and noise

What it does AI-ADS is a curated directory of audio datasets aimed at generative AI and speech research. The repository itself contains no code or data; it is a single README that catalogs external resources—ranging from common ASR corpora like LibriSpeech to niche collections like Genshin Impact voice lines—organized into speech, music, and sound-effect categories. Each entry includes a link and a brief description of what it contains.

The interesting bit The value here is entirely in the curation, which is the boring part most people skip. By treating dataset discovery as a first-class problem, the list saves researchers from the archaeological dig through OpenSLR, Zenodo, and Hugging Face. It even surfaces unusual resources like Carnatic varnam recordings and emotional speech databases that are easy to miss in general search.

Key highlights

• Covers speech, music, and sound effects in separate sections.
• Surfaces specialized resources: emotional voice conversion (ESD), multilingual speech (Emilia, CoVoST), and game audio (Genshin datasets).
• Pure reference material: no code, no scripts, just annotated links.
• 949 stars suggest it fills a genuine discovery gap for audio researchers.

Verdict Audio ML researchers and hobbyists who need a quick survey of training data should bookmark this. If you are looking for a framework, preprocessing pipeline, or hosted data, this is just a table of contents.