LibriSpeech ASR: 1,000 Hours of Aligned English Audiobook Speech

LibriSpeech contains 1,000 hours of 16kHz read English speech derived from LibriVox audiobooks, prepared by Vassil Panayotov and Daniel Povey. The corpus features segmented and aligned audio paired with corresponding text transcripts for speech recognition and speaker identification tasks. The dataset is organized into subsets based on the difficulty of the speech recognition task and the quality of the recordings.

Use Cases

• Training automatic speech recognition (ASR) models using the 16kHz audio and text transcript pairs
• Speaker identification tasks using the provided speaker labels
• Audio classification based on the read audiobook segments

Strengths

• 1,000 hours of 16kHz audio recordings
• Expert-generated alignments between speech and text
• Large scale with 100,000 to 1,000,000 records

Limitations

• Domain bias toward read audiobook speech rather than natural conversation
• Monolingual English coverage only

Provenance

Source

LibriVox project via OpenSLR

Collection Method

Audiobooks were segmented and aligned with text by researchers.

Freshness

Last updated July 2025.