WildVid-LIP: Over 64,000 Temporal Video Segments for Lip Reading

Over 64,000 curated temporal segments from unconstrained, real-world YouTube videos. WildVid-LIP is a large-scale, open-source dataset providing precise timestamp anchors for training Visual Speech Recognition and multimodal models. The dataset was created by Rizul2159 and was last updated on June 16, 2026.

Use Cases

• Train Visual Speech Recognition (VSR) models based on precise timestamp anchors for lip movements.
• Develop audio-visual synchronization models based on aligned video and audio segments.
• Pretrain multimodal self-supervised models based on unconstrained, real-world video content.
• Benchmark lip-reading algorithms based on a large-scale, open-source dataset.

Strengths

• Contains over 64,000 curated temporal segments.
• Focuses on unconstrained, real-world video content from YouTube.
• Provides precise timestamp anchors optimized for specific training tasks.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic, temporal, or content bias inherent to its YouTube source.

Provenance

Source

Curated from YouTube videos.

Collection Method

Likely involves extracting and annotating temporal segments from publicly available videos.

Freshness

Last updated 2026-06-16 16:02:13; freshness should be verified.