nyuuzyou/asmr

Dataset Card for ASMR Audio Dataset

Dataset Summary

This dataset contains a large collection of ASMR (Autonomous Sensory Meridian Response) audio clips with corresponding machine-generated transcriptions. The dataset includes approximately 283,132 audio segments totaling over 307 hours of content, with an average duration of 3.92 seconds per clip. All audio files are provided in WAV format at 24 kHz sampling rate, making them suitable for various audio processing and machine learning applications including automatic speech recognition, text-to-speech synthesis, and audio classification tasks.

The dataset captures the distinctive characteristics of ASMR content, featuring soft-spoken dialogue, whispered conversations, and gentle vocalizations typically associated with ASMR videos and audio content. All clips in the dataset contain speech or vocal content - segments without any speech were filtered out during processing. While the dataset was not filtered by gender, the majority of audio clips feature female voices, which is representative of the broader ASMR content landscape.

Languages

The dataset is primarily monolingual:

• —English (en): Audio content and transcriptions are predominantly in English.

Dataset Structure

Data Files

The dataset consists of:

• —Audio files stored in compressed tar archive (dataset.tar.zst)
• —Metadata file (meta.csv) containing transcriptions and file mappings
• —283,132 individual audio clips in WAV format
• —All audio normalized to 24,000 Hz sampling rate

Data Fields

Each audio clip in the dataset is accompanied by metadata containing the following fields:

• —filename: Unique identifier for the audio file (UUID format without extension)
• —text: Machine-generated transcription of the audio clip using distil-whisper/distil-large-v3.5

Audio Specifications

• —Format: WAV (uncompressed)
• —Sampling Rate: 24,000 Hz
• —Total Files: 283,132
• —Total Duration: 307 hours, 55 minutes, 43.10 seconds (1,108,543.10 seconds)
• —Average Duration: 3.92 seconds per file
• —File Naming: UUID-based filenames (e.g., 633b0111-58a8-4873-bc0d-e40f2c9ba2da.wav)

Data Splits

The dataset contains a single split:

Total dataset size: 283,132 audio-text pairs