schibsted/WAAS

What it solves

WAAS provides a user-friendly way to deploy and use OpenAI Whisper for audio and video transcription. It transforms the raw Whisper model into a complete service by adding a graphical user interface (GUI), a REST API, and an asynchronous job queue, removing the need for users to interact with the model via command line or complex scripts.

How it works

The system consists of three main components: a Flask-based API that handles requests, a Redis-backed job queue (using RQ) to manage transcription tasks, and a worker that processes the audio files using Whisper. Users can submit files via the GUI (called Jojo) or the API, specifying parameters like language and model size. Once processed, the system notifies the user via email or a webhook and provides download links for the results in various formats (SRT, VTT, JSON, or plain text).

Who it’s for

It is designed for people who need to transcribe audio or video files at scale or integrate transcription capabilities into other applications without managing the underlying ML infrastructure manually.

Highlights

• Integrated Editor: Includes a browser-based editor (Jojo) for listening to audio segments and correcting transcription errors locally.
• Asynchronous Processing: Uses a queue system to handle multiple transcription jobs without blocking the API.
• Multiple Output Formats: Supports exporting transcriptions as SRT, VTT, JSON, and plain text.
• Flexible Notifications: Notifies users of completed jobs through either email callbacks or secure webhooks.
• GPU Acceleration: Supports NVIDIA CUDA for faster transcription via Docker.