EMOVA-Alignment-7M: Multimodal Pre-training Data for Vision, Speech, and Language

EMOVA-Alignment-7M is a dataset curated for omni-modal pre-training, including vision-language and speech-language alignment. It was created by Emova-ollm using open-sourced image-text pre-training datasets, OCR datasets, and 2,000 hours of ASR and TTS data. The dataset page was last updated on 2025-03-14.

Use Cases

• Train vision-language models based on the described image-text pre-training data.
• Develop speech-language alignment models based on the 2,000 hours of ASR and TTS data.
• Fine-tune OCR systems using the described OCR datasets.
• Conduct multimodal pre-training research combining visual, textual, and auditory modalities.
• Benchmark omni-modal AI systems on tasks requiring cross-modal understanding.

Strengths

• Integrates 2,000 hours of ASR and TTS data for speech-language alignment.
• Combines multiple data sources including image-text and OCR datasets for multimodal coverage.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and total dataset size are unknown, which may limit suitability assessment.
• Data may reflect source bias inherent to the aggregated open-sourced datasets.

Provenance

Source

Emova-ollm

Collection Method

Curated from open-sourced image-text pre-training datasets, OCR datasets, and ASR/TTS data.

Freshness

Last updated 2025-03-14 13:21:17; freshness should be verified.