Dataset · aoxo
English female ASMR audio corpus with word timestamps
aoxo/t2a-mommy
Female ASMR voice recordings in AAC format with Whisper word-level alignment and non-speech sound tags for ASR and generation tasks.
What's inside
An English audio corpus for the text2asmr project, structured by author. Inside are original audio files in M4A format (48 kHz AAC), JSON files with word-level alignment from Whisper large-v3, and JSONL annotations of non-speech sounds in pauses obtained using Qwen3-Omni.
What you can do with it
- Training and adapting automatic speech recognition (ASR) models on whispering and quiet ASMR audio
- Training audio event classifiers and non-speech sound detectors based on JSONL annotations
- Training text-to-speech and ASMR audio generation models (text-to-audio)
- Tasks
- Audio classification, Automatic speech recognition
- Languages
- English
- Downloads this month
- 23K
- Likes
- 1
How to load it
from datasets import load_dataset
ds = load_dataset("aoxo/t2a-mommy")