Rescript

Product Introduction

1. Definition: Rescript is an open-source, transcript-based video editor (a browser-based, client-side video editing application). It operates as a local-first, offline-capable web app that leverages AI for automatic speech recognition (ASR) and speaker diarization.
2. Core Value Proposition: It exists to provide a private, accessible, and intuitive method for video editing by transforming the process into a text-based workflow. Its primary value is enabling offline browser video editing with word-level precision cuts directly from a transcript, eliminating the need for complex timeline manipulation or cloud uploads.

Main Features

1. Local, Offline-First Processing: All media processing, including AI-powered transcription and final video export, occurs entirely within the user's web browser. The application uses IndexedDB for caching AI models (~90 MB total from Hugging Face Hub) after the first use, ensuring full functionality without an internet connection. This is achieved through technologies like ffmpeg.wasm for media decoding/encoding and ONNX runtime via transformers.js for AI inference.
2. Word-Level Transcript Editing: The core editing mechanic. The application generates a transcript with per-word timestamps using a timestamped Whisper model (whisper-base_timestamped). Users edit video by deleting words or sentences from this transcript text. The corresponding audio and video segments are automatically marked for removal, creating frame-accurate cut points without manual clipping on a timeline.
3. Automated Speaker Diarization & Filler Word Removal: Beyond transcription, the app integrates the pyannote-segmentation-3.0 model to label each word with a speaker identifier, organizing the transcript by speaker. It also includes a one-click feature to automatically detect and remove common filler words ("um," "uh"), leveraging the precise word timestamps for batch editing.
4. In-Browser MP4 Export: The final edited video is rendered and exported as a standard MP4 file (H.264/AAC) directly in the browser. Using a multi-threaded ffmpeg.wasm instance, the application constructs a filter graph to concatenate the non-cut ranges of the original video, ensuring the exported file reflects all transcript edits accurately.

Problems Solved

1. Pain Point: The complexity and learning curve of traditional non-linear video editing software (NLEs) for simple cutting tasks, and the privacy concerns of uploading sensitive video content to cloud-based transcription or editing services.
2. Target Audience: Content creators (podcasters, educators, YouTubers), journalists, researchers, and professionals who need to quickly edit interview footage, lecture recordings, or meeting videos without advanced editing skills. It is particularly valuable for users handling confidential material or with limited internet access.
3. Use Cases: Rapidly editing podcast interviews by removing mistakes and pauses; creating concise highlight reels from long webinar recordings; preparing subtitled video clips for social media by editing the transcript; cleaning up verbal filler words from presentation videos offline.

Unique Advantages

1. Differentiation: Unlike cloud-based transcript editors (Descript, Otter.ai), Rescript processes everything locally, ensuring complete data privacy. Unlike traditional browser video trimmers, it offers AI-powered, word-accurate editing via text. Unlike desktop software, it requires no installation and runs on any modern browser.
2. Key Innovation: The integration of a full, local AI inference pipeline (Whisper + Pyannote) within a browser context using WebGPU/WASM, coupled with a text-centric editing interface that directly manipulates the media timeline. The use of ffmpeg.wasm for client-side, frame-accurate video export completes a truly standalone, professional-grade editing suite in the browser.

Frequently Asked Questions (FAQ)

1. Is Rescript really free and fully offline? Yes, Rescript is a free, open-source (MIT licensed) web application. After the initial one-time download of the AI models, all transcription, editing, and video export processes happen 100% offline in your browser, with no data sent to external servers.
2. What video formats and browsers are supported? Rescript can process most common video formats supported by ffmpeg.wasm (like MP4, MOV, WEBM). For full functionality, a Chromium-based browser (Chrome, Edge, Brave) is recommended as it requires SharedArrayBuffer support and can utilize WebGPU for faster AI transcription.
3. How accurate is the automatic transcription and speaker detection? Accuracy depends on the whisper-base model and audio quality. It is suitable for clear speech but may have errors with heavy accents, background noise, or overlapping talkers. Speaker diarization effectiveness can vary with the number of speakers and audio clarity.
4. Can I edit the video visually on a timeline, not just the text? Rescript's primary interface is the transcript. However, it includes a supplemental timeline view that displays the audio waveform, word segments, cut regions (in red), and a playhead for visual reference and playback control.
5. Where does my video data go when I use Rescript? Your video data never leaves your computer. All processing—audio extraction, AI transcription, editing logic, and final rendering—is executed locally within your browser's sandboxed environment using Web Assembly (WASM) technologies.