Dataset · alexwengg
Speech-text alignment benchmark based on Buckeye corpus
alexwengg/buckeye
16.1 hours of conversational speech with manual word-level time stamps for forced alignment.
What's inside
The dataset is based on the Buckeye Corpus (v2.0) and includes 2,478 audio segments from 20 speakers with a total duration of 16.1 hours (145,762 words, averaging 59 words per segment). The audio is provided in 16 kHz mono PCM WAV format with manual word-level boundary alignments.
What you can do with it
- Benchmarking audio-text forced alignment systems
- Evaluating word timestamp prediction accuracy in ASR models
- Fine-tuning spontaneous conversational speech segmentation algorithms
- Size
- 1K–10K rows
- Downloads this month
- 21.4K
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("alexwengg/buckeye")