nickwebt1/instascript-youtube-transcript-measurements

InstaScript YouTube transcript measurements

Summary

This dataset preserves measured observations from browser-based caption reading, transcript-panel display, transcript download, and Whisper tiny.en browser transcription. The files are the original measurements and the page implementation used for the relevant copy and byte-size behavior; no values have been synthesized or averaged.

Files

• —caption-sample-and-panel-counts.md — caption-track sample rows, transcript-panel line counts, character counts, and lookup times.
• —txt-srt-file-size-measurements.md — measured TXT/SRT byte sizes, cue counts, and download timings.
• —whisper-tiny-en-browser-benchmark.csv — raw Whisper tiny.en browser timing rows, input file bytes, model-load flag, and output character counts.
• —whisper-tiny-en-browser-benchmark-method.md — benchmark method, environment, reproduction steps, and stated limitations.
• —youtube-transcript.js — transcript-page implementation, including panel rendering and copy behavior.
• —youtube-transcript-download.js — download-page implementation, including TXT/SRT construction and Blob byte-size calculation.

Measurement method

The caption and download observations are one run per row on the live page in a real browser on 2026-09-17. Caption-page timing covers lookup, the browser check, and caption download. Download-page file sizes are read from the files that landed on disk. The source notes retain the exact sample names, line counts, character counts, byte counts, and timings.

The Whisper benchmark was measured on 2026-09-14. The timer starts when the Transcribe button is clicked and stops when the transcript panel is visible and contains non-empty text. It covers audio decode, 30-second chunk passes, and rendering; file selection is outside the timer. Run 1 includes model download and initialization, while runs 2–5 reuse the loaded model.

Environment

• —Caption and download pages: the author's laptop and connection, with the exact run details recorded in the two source Markdown files.
• —Whisper benchmark: Microsoft Edge (Chromium), headless, 1280x900 viewport; Windows 11 build 26200; Intel Core Family 6, Model 158; transformers.js/WASM, quantised q8 onnx-community/whisper-tiny.en; measured through a residential connection using a local HTTP proxy.

Limitations

These are single-run observations, not averages or confidence intervals. They describe one machine, one browser, and one network path. Cache state affects lookup and model-load timings; audio content and device load can affect transcription speed. The source notes explicitly mark derived calculations and do not claim an unmeasured accuracy, quota, length ceiling, or file-size ceiling.

License

The dataset is released under the Creative Commons Attribution 4.0 International license (CC BY 4.0). Please retain attribution when reusing it.

Project page: InstaScript