Skip to content

Dataset · ai-sage

Timestamped English speech for temporal audio understanding

ai-sage/TimeGround-1M

Hundreds of thousands of audio clips from 14K hours of speech with timestamp annotations and summaries.

What's inside

A multimodal dataset of 100K to 1M English examples, built from 14K hours of YODAS2 speech corpus data after language and silence filtering. It contains synthetic annotations for temporal localization, timestamp-based event description, and time-aligned summarization. The annotations underwent multi-stage verification via LLMs and deterministic validators.

What you can do with it

  • Train multimodal audio models (Audio-LLM) to answer questions with timestamps
  • Train summarization systems for long audio recordings and podcasts with temporal alignment
Tasks
Question answering, Summarization
Languages
English
Size
100K–1M rows
Downloads this month
25.9K
Likes
8

How to load it

from datasets import load_dataset

ds = load_dataset("ai-sage/TimeGround-1M")