1,000 hours of audio recordings and transcriptions derived from LibriVox and Project Gutenberg for speech recognition and synthesis. The collection features French audio clips between 1 and 20 seconds in length paired with literary texts published from 1884 to 1964.
Use Cases
- Train speech synthesis models using the audio clips and matching transcriptions.
- Fine-tune speech-to-text engines using the prepared text-files as ground truth labels.
- Analyze phonetic variations in French literary readings using the clip-level transcriptions.
Strengths
- Nearly 1,000 hours of total audio data across the M-AILABS collection.
- Audio clips standardized to lengths between 1 and 20 seconds.
- Includes detailed metadata for each subset within info.txt files.
- Text transcriptions provided for every individual audio clip.