How jack-zc8/m3av-dataset works
M³AV is an academic-lecture dataset release with sample data and benchmark implementations. Its core workflow is to provide synchronized speech, slide/OCR and (for some computer-science videos) paper data for research; the demo script additionally aligns slide OCR text with timestamped spoken words. Benchmark branches cover ASR, contextual ASR, summarization and spontaneous TTS. The README establishes the dataset modalities and benchmark scope; only the demo alignment script was sampled, so other benchmark internals and their wiring are represented conservatively.
Main entry points: get_seg_speech_ocr.py OCR-speech aligner · speech Timestamped speech · ocr Slide OCR · download Download release
Components 15
Dataset release
- Download release
download - Timestamped speech
demo/speech - Slide OCR
demo/ocr - Paper sentences
demo/paper - Video reference
demo/raw
Demo alignment
- OCR-speech aligner
demo/get_seg_speech_ocr.py - Segmented OCR speech
demo/ocr
Research benchmarks
- ASR benchmark
benchmarks/ASR - Contextual ASR
benchmarks/CASR - SSG dataset tools
benchmarks/SSG/get_ssg_dataset - SSG generation
benchmarks/SSG - SSG scoring
benchmarks/SSG/score - Spontaneous TTS
benchmarks/Spontaneous_TTS
Other components
- Researcher
- Lecture videos
Connections 14
- Researcher → Download release: downloads
- Lecture videos → Video reference: references
- Researcher → Timestamped speech: uses
- Researcher → Slide OCR: uses
- Researcher → Paper sentences: uses
- Slide OCR → OCR-speech aligner: reads OCR
- Timestamped speech → OCR-speech aligner: reads timestamps
- OCR-speech aligner → Segmented OCR speech: writes segments
- Researcher → ASR benchmark: runs benchmark
- Researcher → Contextual ASR: runs benchmark
- Researcher → SSG dataset tools: builds data
- Researcher → SSG generation: runs generation
- Researcher → SSG scoring: evaluates output
- Researcher → Spontaneous TTS: runs benchmark