Annotated Audio Dataset of Bangladeshi English Speakers for Pause-Based Fluency Analysis

This dataset contains 57 audio recordings of spoken English collected for the purpose of studying oral fluency, specifically through the analysis of filled and silent pauses. The speakers are Bangladeshi English speakers, and the recordings represent various fluency levels.

• The audio files are stored in the data/ folder and are available in multiple formats, including .mp3, .wav, and .m4a.
• Each audio file has a corresponding annotation file in JSON format, located in the JSON/ folder.
• The JSON files include detailed time-stamped annotations of pause markers (both filled and silent), speaker metadata, and fluency-related labels used for machine learning tasks.

This dataset is intended for use in speech processing, NLP, language learning research, and machine learning applications related to fluency assessment. It has been used in research involving transformer-based models, pause detection, and low-resource learning scenarios.