# Caption Review > Compare an existing English caption track with its recording, review disagreements and export deliberate edits. This is a personal software project by Prateek Mulye. It is free to use without an account. It does not represent an employer deployment or customer adoption. ## Workspace and source - [Open Caption Review](https://captions.prateekmulye.dev/) - [Source code](https://github.com/prateekmulye/caption-review) - [Prateek Mulye's portfolio](https://prateekmulye.dev/) ## What it does Inputs: A local recording and a plain-text SRT or WebVTT caption track. Up to 10 minutes of supported media and 500 cues, subject to browser decoding limits. AI contribution: Xenova/whisper-tiny.en speech recognition runs locally with Transformers.js. Public model files download from pinned Hugging Face assets after an explicit action. Outputs: Revised SRT or WebVTT, original captions and JSON edit log. Pending or rejected suggestions do not enter the revised track. ## Review and limitations AI comparison requires supported mono or stereo PCM WAV within the displayed duration and decoded-memory limits. Other supported media formats allow playback and manual editing. Speech recognition can mishear names, accents and overlapping speech. No transcription-accuracy, accessibility-certification or time-saving guarantee. A person reviews the findings. The application does not treat model output as an authorized external action. Functional checks and synthetic examples do not establish operational accuracy or business impact. ## Data handling Input contents, decisions and exports are processed in the browser. The app does not upload input contents or persist review sessions across reloads. Public runtime and model files are downloaded only when the user starts AI analysis; hosting providers can receive ordinary request metadata. Public model assets may remain cached. No account, analytics or input-content telemetry is required. ## Attribution Author: Prateek Mulye, Senior Software Engineer and applied AI practitioner.