LibriSpeech English Speech Corpus for ASR Testing

The LibriSpeech ASR Test dataset contains approximately 1000 hours of 16 kHz English speech derived from LibriVox audiobooks. It was prepared by Vassil Panayotov with assistance from Daniel Povey and is carefully segmented and aligned.

Use Cases

• Train and evaluate automatic speech recognition (ASR) models on 1000 hours of segmented English speech.
• Benchmark ASR system performance on read audiobook speech with aligned transcriptions.
• Analyze acoustic features from 16 kHz audio files for speech processing research.

Strengths

• Approximately 1000 hours of speech data provides substantial material for model training.
• Audio is carefully segmented and aligned, facilitating direct use for ASR tasks.
• Recordings are in 16 kHz FLAC format, a standard for speech processing.

Limitations

• The dataset consists solely of read speech from audiobooks, which may not represent conversational or noisy speech patterns.
• Audio is stored in FLAC format and requires conversion to float32 arrays for typical ML workflows, adding a preprocessing step.

Provenance

Source

LibriVox project audiobooks.

Collection Method

Derived from read audiobooks, carefully segmented and aligned.

Time Range

null

Freshness

null

Geography

null