Unverified60% confidenceFactExact time
Whisper's training data contains 680,000 hours, largely sourced from videos on the internet with varying subtitle quality.
2
Sources
60%
Confidence
Long-term
Relevance
7/22/2026
First Seen
Sources
Related Claims
Unverified训练使用约500万行英文维基百科语料约3亿token，10万上下文词表，23758个名词目标67% similarUnverifiedCLIP的训练数据来自互联网公开爬取的数据集WIT（WebImageText），其规模约为ImageNet（约1400万张标注图像）的近30倍65% similarUnverifiedCommon Crawl是非营利组织运营的大规模网络爬取项目，每月抓取数十亿网页，数据集被用于训练包括GPT系列在内的大型语言模型64% similarUnverifiedThe training data for modern image generation models typically comes from billions of internet images and their accompanying text descriptions.63% similarUnverified卡帕西在YouTube上发布的神经网络系列教程累计播放量超过千万63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/587207API
curl https://kongchang.com/api/v1/knowledge/claims/587207MCP
get_claim(id=587207)