Whisper with a queue, email, and a browser editor

What it does WAAS is a service layer over OpenAI Whisper. A Flask API backed by Redis and RQ accepts audio or video files, queues them for transcription or translation, and returns a job_id to track progress. A web frontend called Jojo handles uploads directly, then emails download links for SRT, VTT, plain text, or a custom Jojo file. There is also a browser-based editor that runs entirely offline to review segments, listen back, and fix transcription errors.

The interesting bit The editor is the unusual part: it works 100% in the browser with no server round-trips, turning the browser into a local transcription correction desk. The project also bakes in newsroom pragmatism—born at Norwegian newspaper VG, it requires either an email address or a webhook for every job, so it is built around notification rather than idle polling.

Key highlights

• Async job queue via Redis/RQ with support for multiple Whisper models (defaulting to tiny).
• Browser-based “Jojo” editor for segment-by-segment correction, saving work locally to a custom file format.
• Webhook callbacks with X-WAAS-Signature headers for verification, plus email delivery of results.
• GPU acceleration available via an alternative Docker setup for NVIDIA CUDA.
• Output formats include SRT, VTT, plain text, timecoded text, and raw JSON.

Caveats

• The README has visible copy-paste errors and typos (e.g., “asynchroniously,” SRT described as a WebVTT file), so the project feels maintained but not meticulously polished.
• It is fundamentally glue code—Flask, Redis, RQ, and a Whisper wrapper—so expect to configure and host it yourself.
• Python 3.8–3.10 compatibility and VRAM requirements are inherited directly from upstream Whisper.

Verdict Newsrooms, podcasters, or teams already self-hosting Whisper who need a shared queue and a non-technical correction interface will find this useful. If you just need a quick CLI transcription or an OpenAI API endpoint, this is overkill.