AI · Collection
Audio Datasets - Third Party Classics
The major public speech and audio datasets, kept for reference in case they are used in training.
21 datasets
- dataset openslr/librispeech_asr About 1,000 hours of read English audiobooks; the standard ASR benchmark.
- dataset MLCommons/peoples_speech Very large open English speech corpus.
- dataset speechcolab/gigaspeech 10,000 hours of English audiobooks, podcasts and YouTube. Gated.
- dataset facebook/multilingual_librispeech Read audiobook speech at scale, several languages.
- dataset parler-tts/mls_eng English portion of Multilingual LibriSpeech.
- dataset facebook/voxpopuli European Parliament speech, multilingual.
- dataset google/fleurs Read-speech benchmark across about 100 languages.
- dataset espnet/yodas YouTube-derived multilingual speech at very large scale.
- dataset nvidia/Granary Large multilingual ASR and translation corpus with pseudo-labels.
- dataset edinburghcstr/ami Meeting recordings: multi-speaker, conversational.
- dataset kensho/spgispeech 5,000 hours of earnings calls, formatted transcripts. Gated.
- dataset distil-whisper/earnings22 Accented earnings-call speech; long-form benchmark.
- dataset keithito/lj_speech Classic single-speaker TTS set. Older script loader.
- dataset mythicinfinity/libritts LibriSpeech re-cut for TTS.
- dataset mythicinfinity/libritts_r LibriTTS with restored audio quality.
- dataset CSTR-Edinburgh/vctk About 110 English speakers, varied accents. Older script loader.
- dataset amphion/Emilia-Dataset Large in-the-wild multilingual speech for TTS. Gated.
- dataset google/speech_commands Short spoken keywords, for keyword spotting. Older script loader.
- dataset MLCommons/ml_spoken_words Multilingual spoken keywords.
- dataset agkphysics/AudioSet Sound-event classification, not transcription.
- dataset speechcolab/gigaspeech3 Very large English speech corpus; newer GigaSpeech release.
Copied from Hugging Face on 7 November 2026.