Running Whisper transcription on decentralized compute

[suzado]

Hi everyone,

I’d like to share a project I built as an experiment combining Whisper with decentralized computing.

The idea is to run Whisper transcription tasks on remote providers using the Golem network. The workflow involves sending a media file together with a containerized Whisper model to a selected provider, which performs transcription and returns results in formats such as TXT or SRT with word-level timestamps.

Key aspects:

• containerized Whisper inference running on remote providers
• support for batch processing (multiple providers in parallel)
• word-level synchronized playback UI
• AI-based transcript summarization on top of Whisper output

This setup makes it useful for meeting recordings, voice notes, YouTube material, and longer video content.

From my observations, performance can vary depending on provider availability and network conditions, especially container image upload time.

I built this mainly to better understand Whisper deployment patterns and experiment with decentralized AI workloads.

I’d be happy to hear feedback, suggestions, or thoughts from others experimenting with Whisper in distributed or remote inference setups.

Project is free and open source: https://github.com/suzado/gScribe

Landing Page	Processing	Results