LibriSpeech 1000-Hour English Speech Corpus

LibriSpeech contains approximately 1000 hours of 16kHz read English speech. The corpus was prepared by Vassil Panayotov with assistance from Daniel Povey, derived from audiobooks in the LibriVox project. The dataset was uploaded to Hugging Face by nguyenvulebinh in December 2022.

Use Cases

• Train automatic speech recognition models on 1000 hours of segmented and aligned English speech audio.
• Develop speaker identification systems using audio derived from multiple LibriVox readers.
• Benchmark audio segmentation algorithms on carefully prepared speech segments.
• Fine-tune acoustic models for read English speech using the 16kHz audio samples.

Strengths

• Approximately 1000 hours of audio data provides substantial training material.
• Data is carefully segmented and aligned, ensuring high-quality labels for speech tasks.
• Audio is recorded at a consistent 16kHz sampling rate.

Limitations

• Content is exclusively from read audiobooks, which may not represent conversational or noisy speech.
• The dataset is derived from LibriVox, potentially introducing bias towards specific accents or reading styles.
• Specific details on speaker count, demographic distribution, or class balance are not provided.

Provenance

Source

LibriVox project audiobooks.

Collection Method

Derived from read audiobooks, then segmented and aligned.

Freshness

Last updated on the platform in December 2022.