VoxCeleb
VoxCeleb is a large speaker-recognition dataset built by the Visual Geometry Group at the University of Oxford. It gathers short speech segments from celebrity interviews on YouTube, spanning a wide range of accents, ages, and recording conditions, which makes it a realistic testbed for identifying and verifying who is speaking.
The dataset is a common benchmark for speaker verification (deciding whether two clips are the same person) and speaker identification. Note the split licence: the annotations Oxford provides are released under CC BY 4.0, but the audio itself comes from YouTube videos that remain under their original terms. That means the underlying content carries no blanket reuse right, so commercial use is restricted and you should check the source terms before building a product on it.
Related sources
LibriSpeech
A collection of 1,000 hours of read English speech, released under CC BY 4.0 and stored using the open-source FLAC audio encoder. Labels are aligned at the sentence level. Derived from LibriVox public domain audiobooks, it is the standard benchmark for English automatic speech recognition.
LJ Speech
An open dataset of 13,100 short audio clips of a single speaker reading passages from seven non-fiction books. Every clip is transcribed and clips run from 1 to 10 seconds. Released into the public domain, it is the standard single-speaker text-to-speech benchmark.
Mozilla Common Voice
A crowdsourced speech platform from Mozilla that releases datasets under the CC0 licence. Volunteers read sentences aloud and other community members validate each recording. As of release 19.0 it holds 32,584 hours of speech across 131 languages, making it one of the largest openly licensed voice datasets available.
OpenSLR
Open Speech and Language Resources, a hosting site for speech and language datasets, software, and models. It is the home of LibriSpeech and dozens of other language-specific speech corpora, making it a central catalogue for finding openly available voice data.