license: cc-by-4.0
language:
- en
Dataset Card for youtube-commons-asr-eval
Table of Contents
Dataset Summary
This evaluation dataset is created from a subset of Youtube-Commons [PleIAs/YouTube-Commons] by selecting English YouTube videos and corresponding english subtitle.
Supported Tasks and Leaderboards
This dataset will be primarily useful for automatic speech recognition evaluation tasks such as hf-audio/open_asr_leaderboard.
Languages
This subset is for English language evaluations.
Dataset Structure
The dataset consists of 94 video links, transcriptions, and normalized transcriptions (around 38 hours) of age-appropriate audios with a minimum word count of 300. With a normal speaking rate of 2.5 words per second, this corresponds to a minimum duration of 2 minutes. Minimum duration of the dataset is 128 seconds and maximum is 02:08 hours. The average duration per file is a little over 24 minutes and the standard deviation is 25 minutes. The notable variability in audio duration, as indicated by the standard deviation, mirrors typical real-time environments.
Data Fields
Each row in the JSON file has link (link to the youtube video), text (transcription), norm_text (normalized transcription) and duration (duration of the video) fields.
Data Splits
Evaluation data
Dataset Creation
Normalization is done via EnglishTextNormalizer from open_asr_eval [https://github.com/huggingface/open_asr_leaderboard/blob/main/normalizer/normalizer.py] The dataset is created by selecting the first 100 files from Youtube-Commons, with a minimum of 300 transcription words and age-appropriate content. Three files are manually removed owing to high errors in the transcription observed in visual inspection and also verified with high WER on different ASR implementations.
Licensing Information
All the transcripts are part of a video shared under a CC-By license on YouTube. All the licensing terms are the same as the original dataset [PleIAs/YouTube-Commons].